blob: 87368ae5da24ad768333380ee46a25af9116de60 [file] [log] [blame]
Willy Tarreau2212e6a2015-10-13 14:40:55 +02001 ------------------------
2 HAProxy Management Guide
3 ------------------------
Willy Tarreaufba74ea2018-12-22 11:19:45 +01004 version 2.0
Willy Tarreau2212e6a2015-10-13 14:40:55 +02005
6
7This document describes how to start, stop, manage, and troubleshoot HAProxy,
8as well as some known limitations and traps to avoid. It does not describe how
9to configure it (for this please read configuration.txt).
10
11Note to documentation contributors :
12 This document is formatted with 80 columns per line, with even number of
13 spaces for indentation and without tabs. Please follow these rules strictly
14 so that it remains easily printable everywhere. If you add sections, please
15 update the summary below for easier searching.
16
17
18Summary
19-------
20
211. Prerequisites
222. Quick reminder about HAProxy's architecture
233. Starting HAProxy
244. Stopping and restarting HAProxy
255. File-descriptor limitations
266. Memory management
277. CPU usage
288. Logging
299. Statistics and monitoring
Willy Tarreau44aed902015-10-13 14:45:29 +0200309.1. CSV format
Willy Tarreau5d8b9792016-03-11 11:09:34 +0100319.2. Typed output format
329.3. Unix Socket commands
William Lallemand142db372018-12-11 18:56:45 +0100339.4. Master CLI
Willy Tarreau2212e6a2015-10-13 14:40:55 +02003410. Tricks for easier configuration management
3511. Well-known traps to avoid
3612. Debugging and performance issues
3713. Security considerations
38
39
401. Prerequisites
41----------------
42
43In this document it is assumed that the reader has sufficient administration
44skills on a UNIX-like operating system, uses the shell on a daily basis and is
45familiar with troubleshooting utilities such as strace and tcpdump.
46
47
482. Quick reminder about HAProxy's architecture
49----------------------------------------------
50
Willy Tarreau3f364482019-02-27 15:01:46 +010051HAProxy is a multi-threaded, event-driven, non-blocking daemon. This means is
Willy Tarreau2212e6a2015-10-13 14:40:55 +020052uses event multiplexing to schedule all of its activities instead of relying on
53the system to schedule between multiple activities. Most of the time it runs as
54a single process, so the output of "ps aux" on a system will report only one
55"haproxy" process, unless a soft reload is in progress and an older process is
56finishing its job in parallel to the new one. It is thus always easy to trace
Willy Tarreau3f364482019-02-27 15:01:46 +010057its activity using the strace utility. In order to scale with the number of
58available processors, by default haproxy will start one worker thread per
59processor it is allowed to run on. Unless explicitly configured differently,
60the incoming traffic is spread over all these threads, all running the same
61event loop. A great care is taken to limit inter-thread dependencies to the
62strict minimum, so as to try to achieve near-linear scalability. This has some
63impacts such as the fact that a given connection is served by a single thread.
64Thus in order to use all available processing capacity, it is needed to have at
65least as many connections as there are threads, which is almost always granted.
Willy Tarreau2212e6a2015-10-13 14:40:55 +020066
67HAProxy is designed to isolate itself into a chroot jail during startup, where
68it cannot perform any file-system access at all. This is also true for the
69libraries it depends on (eg: libc, libssl, etc). The immediate effect is that
70a running process will not be able to reload a configuration file to apply
71changes, instead a new process will be started using the updated configuration
72file. Some other less obvious effects are that some timezone files or resolver
73files the libc might attempt to access at run time will not be found, though
74this should generally not happen as they're not needed after startup. A nice
75consequence of this principle is that the HAProxy process is totally stateless,
76and no cleanup is needed after it's killed, so any killing method that works
77will do the right thing.
78
79HAProxy doesn't write log files, but it relies on the standard syslog protocol
80to send logs to a remote server (which is often located on the same system).
81
82HAProxy uses its internal clock to enforce timeouts, that is derived from the
83system's time but where unexpected drift is corrected. This is done by limiting
84the time spent waiting in poll() for an event, and measuring the time it really
85took. In practice it never waits more than one second. This explains why, when
86running strace over a completely idle process, periodic calls to poll() (or any
87of its variants) surrounded by two gettimeofday() calls are noticed. They are
88normal, completely harmless and so cheap that the load they imply is totally
89undetectable at the system scale, so there's nothing abnormal there. Example :
90
91 16:35:40.002320 gettimeofday({1442759740, 2605}, NULL) = 0
92 16:35:40.002942 epoll_wait(0, {}, 200, 1000) = 0
93 16:35:41.007542 gettimeofday({1442759741, 7641}, NULL) = 0
94 16:35:41.007998 gettimeofday({1442759741, 8114}, NULL) = 0
95 16:35:41.008391 epoll_wait(0, {}, 200, 1000) = 0
96 16:35:42.011313 gettimeofday({1442759742, 11411}, NULL) = 0
97
98HAProxy is a TCP proxy, not a router. It deals with established connections that
99have been validated by the kernel, and not with packets of any form nor with
100sockets in other states (eg: no SYN_RECV nor TIME_WAIT), though their existence
101may prevent it from binding a port. It relies on the system to accept incoming
102connections and to initiate outgoing connections. An immediate effect of this is
103that there is no relation between packets observed on the two sides of a
104forwarded connection, which can be of different size, numbers and even family.
105Since a connection may only be accepted from a socket in LISTEN state, all the
106sockets it is listening to are necessarily visible using the "netstat" utility
107to show listening sockets. Example :
108
109 # netstat -ltnp
110 Active Internet connections (only servers)
111 Proto Recv-Q Send-Q Local Address Foreign Address State PID/Program name
112 tcp 0 0 0.0.0.0:22 0.0.0.0:* LISTEN 1629/sshd
113 tcp 0 0 0.0.0.0:80 0.0.0.0:* LISTEN 2847/haproxy
114 tcp 0 0 0.0.0.0:443 0.0.0.0:* LISTEN 2847/haproxy
115
116
1173. Starting HAProxy
118-------------------
119
120HAProxy is started by invoking the "haproxy" program with a number of arguments
121passed on the command line. The actual syntax is :
122
123 $ haproxy [<options>]*
124
125where [<options>]* is any number of options. An option always starts with '-'
126followed by one of more letters, and possibly followed by one or multiple extra
127arguments. Without any option, HAProxy displays the help page with a reminder
128about supported options. Available options may vary slightly based on the
129operating system. A fair number of these options overlap with an equivalent one
130if the "global" section. In this case, the command line always has precedence
131over the configuration file, so that the command line can be used to quickly
132enforce some settings without touching the configuration files. The current
133list of options is :
134
135 -- <cfgfile>* : all the arguments following "--" are paths to configuration
Maxime de Roucy379d9c72016-05-13 23:52:56 +0200136 file/directory to be loaded and processed in the declaration order. It is
137 mostly useful when relying on the shell to load many files that are
138 numerically ordered. See also "-f". The difference between "--" and "-f" is
139 that one "-f" must be placed before each file name, while a single "--" is
140 needed before all file names. Both options can be used together, the
141 command line ordering still applies. When more than one file is specified,
142 each file must start on a section boundary, so the first keyword of each
143 file must be one of "global", "defaults", "peers", "listen", "frontend",
144 "backend", and so on. A file cannot contain just a server list for example.
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200145
Maxime de Roucy379d9c72016-05-13 23:52:56 +0200146 -f <cfgfile|cfgdir> : adds <cfgfile> to the list of configuration files to be
147 loaded. If <cfgdir> is a directory, all the files (and only files) it
Dan Lloyd8e48b872016-07-01 21:01:18 -0400148 contains are added in lexical order (using LC_COLLATE=C) to the list of
Maxime de Roucy379d9c72016-05-13 23:52:56 +0200149 configuration files to be loaded ; only files with ".cfg" extension are
150 added, only non hidden files (not prefixed with ".") are added.
151 Configuration files are loaded and processed in their declaration order.
152 This option may be specified multiple times to load multiple files. See
153 also "--". The difference between "--" and "-f" is that one "-f" must be
154 placed before each file name, while a single "--" is needed before all file
155 names. Both options can be used together, the command line ordering still
156 applies. When more than one file is specified, each file must start on a
157 section boundary, so the first keyword of each file must be one of
158 "global", "defaults", "peers", "listen", "frontend", "backend", and so on.
159 A file cannot contain just a server list for example.
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200160
161 -C <dir> : changes to directory <dir> before loading configuration
162 files. This is useful when using relative paths. Warning when using
163 wildcards after "--" which are in fact replaced by the shell before
164 starting haproxy.
165
166 -D : start as a daemon. The process detaches from the current terminal after
167 forking, and errors are not reported anymore in the terminal. It is
168 equivalent to the "daemon" keyword in the "global" section of the
169 configuration. It is recommended to always force it in any init script so
170 that a faulty configuration doesn't prevent the system from booting.
171
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200172 -L <name> : change the local peer name to <name>, which defaults to the local
William Lallemanddaf4cd22018-04-17 16:46:13 +0200173 hostname. This is used only with peers replication. You can use the
174 variable $HAPROXY_LOCALPEER in the configuration file to reference the
175 peer name.
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200176
177 -N <limit> : sets the default per-proxy maxconn to <limit> instead of the
178 builtin default value (usually 2000). Only useful for debugging.
179
180 -V : enable verbose mode (disables quiet mode). Reverts the effect of "-q" or
181 "quiet".
182
William Lallemande202b1e2017-06-01 17:38:56 +0200183 -W : master-worker mode. It is equivalent to the "master-worker" keyword in
184 the "global" section of the configuration. This mode will launch a "master"
185 which will monitor the "workers". Using this mode, you can reload HAProxy
186 directly by sending a SIGUSR2 signal to the master. The master-worker mode
187 is compatible either with the foreground or daemon mode. It is
188 recommended to use this mode with multiprocess and systemd.
189
Pavlos Parissisf65f2572018-02-07 21:42:16 +0100190 -Ws : master-worker mode with support of `notify` type of systemd service.
191 This option is only available when HAProxy was built with `USE_SYSTEMD`
192 build option enabled.
193
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200194 -c : only performs a check of the configuration files and exits before trying
195 to bind. The exit status is zero if everything is OK, or non-zero if an
196 error is encountered.
197
198 -d : enable debug mode. This disables daemon mode, forces the process to stay
199 in foreground and to show incoming and outgoing events. It is equivalent to
200 the "global" section's "debug" keyword. It must never be used in an init
201 script.
202
203 -dG : disable use of getaddrinfo() to resolve host names into addresses. It
204 can be used when suspecting that getaddrinfo() doesn't work as expected.
205 This option was made available because many bogus implementations of
206 getaddrinfo() exist on various systems and cause anomalies that are
207 difficult to troubleshoot.
208
Dan Lloyd8e48b872016-07-01 21:01:18 -0400209 -dM[<byte>] : forces memory poisoning, which means that each and every
Willy Tarreaubafbe012017-11-24 17:34:44 +0100210 memory region allocated with malloc() or pool_alloc() will be filled with
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200211 <byte> before being passed to the caller. When <byte> is not specified, it
212 defaults to 0x50 ('P'). While this slightly slows down operations, it is
213 useful to reliably trigger issues resulting from missing initializations in
214 the code that cause random crashes. Note that -dM0 has the effect of
215 turning any malloc() into a calloc(). In any case if a bug appears or
216 disappears when using this option it means there is a bug in haproxy, so
217 please report it.
218
219 -dS : disable use of the splice() system call. It is equivalent to the
220 "global" section's "nosplice" keyword. This may be used when splice() is
221 suspected to behave improperly or to cause performance issues, or when
222 using strace to see the forwarded data (which do not appear when using
223 splice()).
224
225 -dV : disable SSL verify on the server side. It is equivalent to having
226 "ssl-server-verify none" in the "global" section. This is useful when
227 trying to reproduce production issues out of the production
228 environment. Never use this in an init script as it degrades SSL security
229 to the servers.
230
231 -db : disable background mode and multi-process mode. The process remains in
232 foreground. It is mainly used during development or during small tests, as
233 Ctrl-C is enough to stop the process. Never use it in an init script.
234
235 -de : disable the use of the "epoll" poller. It is equivalent to the "global"
236 section's keyword "noepoll". It is mostly useful when suspecting a bug
237 related to this poller. On systems supporting epoll, the fallback will
238 generally be the "poll" poller.
239
240 -dk : disable the use of the "kqueue" poller. It is equivalent to the
241 "global" section's keyword "nokqueue". It is mostly useful when suspecting
242 a bug related to this poller. On systems supporting kqueue, the fallback
243 will generally be the "poll" poller.
244
245 -dp : disable the use of the "poll" poller. It is equivalent to the "global"
246 section's keyword "nopoll". It is mostly useful when suspecting a bug
247 related to this poller. On systems supporting poll, the fallback will
248 generally be the "select" poller, which cannot be disabled and is limited
249 to 1024 file descriptors.
250
Willy Tarreau3eed10e2016-11-07 21:03:16 +0100251 -dr : ignore server address resolution failures. It is very common when
252 validating a configuration out of production not to have access to the same
253 resolvers and to fail on server address resolution, making it difficult to
254 test a configuration. This option simply appends the "none" method to the
255 list of address resolution methods for all servers, ensuring that even if
256 the libc fails to resolve an address, the startup sequence is not
257 interrupted.
258
Willy Tarreau70060452015-12-14 12:46:07 +0100259 -m <limit> : limit the total allocatable memory to <limit> megabytes across
260 all processes. This may cause some connection refusals or some slowdowns
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200261 depending on the amount of memory needed for normal operations. This is
Willy Tarreau70060452015-12-14 12:46:07 +0100262 mostly used to force the processes to work in a constrained resource usage
263 scenario. It is important to note that the memory is not shared between
264 processes, so in a multi-process scenario, this value is first divided by
265 global.nbproc before forking.
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200266
267 -n <limit> : limits the per-process connection limit to <limit>. This is
268 equivalent to the global section's keyword "maxconn". It has precedence
269 over this keyword. This may be used to quickly force lower limits to avoid
270 a service outage on systems where resource limits are too low.
271
272 -p <file> : write all processes' pids into <file> during startup. This is
273 equivalent to the "global" section's keyword "pidfile". The file is opened
274 before entering the chroot jail, and after doing the chdir() implied by
275 "-C". Each pid appears on its own line.
276
277 -q : set "quiet" mode. This disables some messages during the configuration
278 parsing and during startup. It can be used in combination with "-c" to
279 just check if a configuration file is valid or not.
280
William Lallemand142db372018-12-11 18:56:45 +0100281 -S <bind>[,bind_options...]: in master-worker mode, bind a master CLI, which
282 allows the access to every processes, running or leaving ones.
283 For security reasons, it is recommended to bind the master CLI to a local
284 UNIX socket. The bind options are the same as the keyword "bind" in
285 the configuration file with words separated by commas instead of spaces.
286
287 Note that this socket can't be used to retrieve the listening sockets from
288 an old process during a seamless reload.
289
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200290 -sf <pid>* : send the "finish" signal (SIGUSR1) to older processes after boot
291 completion to ask them to finish what they are doing and to leave. <pid>
292 is a list of pids to signal (one per argument). The list ends on any
293 option starting with a "-". It is not a problem if the list of pids is
294 empty, so that it can be built on the fly based on the result of a command
295 like "pidof" or "pgrep".
296
297 -st <pid>* : send the "terminate" signal (SIGTERM) to older processes after
298 boot completion to terminate them immediately without finishing what they
299 were doing. <pid> is a list of pids to signal (one per argument). The list
300 is ends on any option starting with a "-". It is not a problem if the list
301 of pids is empty, so that it can be built on the fly based on the result of
302 a command like "pidof" or "pgrep".
303
304 -v : report the version and build date.
305
306 -vv : display the version, build options, libraries versions and usable
307 pollers. This output is systematically requested when filing a bug report.
308
Olivier Houchardd33fc3a2017-04-05 22:50:59 +0200309 -x <unix_socket> : connect to the specified socket and try to retrieve any
310 listening sockets from the old process, and use them instead of trying to
311 bind new ones. This is useful to avoid missing any new connection when
William Lallemandf6975e92017-05-26 17:42:10 +0200312 reloading the configuration on Linux. The capability must be enable on the
313 stats socket using "expose-fd listeners" in your configuration.
Olivier Houchardd33fc3a2017-04-05 22:50:59 +0200314
Dan Lloyd8e48b872016-07-01 21:01:18 -0400315A safe way to start HAProxy from an init file consists in forcing the daemon
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200316mode, storing existing pids to a pid file and using this pid file to notify
317older processes to finish before leaving :
318
319 haproxy -f /etc/haproxy.cfg \
320 -D -p /var/run/haproxy.pid -sf $(cat /var/run/haproxy.pid)
321
322When the configuration is split into a few specific files (eg: tcp vs http),
323it is recommended to use the "-f" option :
324
325 haproxy -f /etc/haproxy/global.cfg -f /etc/haproxy/stats.cfg \
326 -f /etc/haproxy/default-tcp.cfg -f /etc/haproxy/tcp.cfg \
327 -f /etc/haproxy/default-http.cfg -f /etc/haproxy/http.cfg \
328 -D -p /var/run/haproxy.pid -sf $(cat /var/run/haproxy.pid)
329
330When an unknown number of files is expected, such as customer-specific files,
331it is recommended to assign them a name starting with a fixed-size sequence
332number and to use "--" to load them, possibly after loading some defaults :
333
334 haproxy -f /etc/haproxy/global.cfg -f /etc/haproxy/stats.cfg \
335 -f /etc/haproxy/default-tcp.cfg -f /etc/haproxy/tcp.cfg \
336 -f /etc/haproxy/default-http.cfg -f /etc/haproxy/http.cfg \
337 -D -p /var/run/haproxy.pid -sf $(cat /var/run/haproxy.pid) \
338 -f /etc/haproxy/default-customers.cfg -- /etc/haproxy/customers/*
339
340Sometimes a failure to start may happen for whatever reason. Then it is
341important to verify if the version of HAProxy you are invoking is the expected
342version and if it supports the features you are expecting (eg: SSL, PCRE,
343compression, Lua, etc). This can be verified using "haproxy -vv". Some
344important information such as certain build options, the target system and
345the versions of the libraries being used are reported there. It is also what
346you will systematically be asked for when posting a bug report :
347
348 $ haproxy -vv
349 HA-Proxy version 1.6-dev7-a088d3-4 2015/10/08
350 Copyright 2000-2015 Willy Tarreau <willy@haproxy.org>
351
352 Build options :
353 TARGET = linux2628
354 CPU = generic
355 CC = gcc
356 CFLAGS = -pg -O0 -g -fno-strict-aliasing -Wdeclaration-after-statement \
357 -DBUFSIZE=8030 -DMAXREWRITE=1030 -DSO_MARK=36 -DTCP_REPAIR=19
358 OPTIONS = USE_ZLIB=1 USE_DLMALLOC=1 USE_OPENSSL=1 USE_LUA=1 USE_PCRE=1
359
360 Default settings :
361 maxconn = 2000, bufsize = 8030, maxrewrite = 1030, maxpollevents = 200
362
363 Encrypted password support via crypt(3): yes
364 Built with zlib version : 1.2.6
365 Compression algorithms supported : identity("identity"), deflate("deflate"), \
366 raw-deflate("deflate"), gzip("gzip")
367 Built with OpenSSL version : OpenSSL 1.0.1o 12 Jun 2015
368 Running on OpenSSL version : OpenSSL 1.0.1o 12 Jun 2015
369 OpenSSL library supports TLS extensions : yes
370 OpenSSL library supports SNI : yes
371 OpenSSL library supports prefer-server-ciphers : yes
372 Built with PCRE version : 8.12 2011-01-15
373 PCRE library supports JIT : no (USE_PCRE_JIT not set)
374 Built with Lua version : Lua 5.3.1
375 Built with transparent proxy support using: IP_TRANSPARENT IP_FREEBIND
376
377 Available polling systems :
378 epoll : pref=300, test result OK
379 poll : pref=200, test result OK
380 select : pref=150, test result OK
381 Total: 3 (3 usable), will use epoll.
382
383The relevant information that many non-developer users can verify here are :
384 - the version : 1.6-dev7-a088d3-4 above means the code is currently at commit
385 ID "a088d3" which is the 4th one after after official version "1.6-dev7".
386 Version 1.6-dev7 would show as "1.6-dev7-8c1ad7". What matters here is in
387 fact "1.6-dev7". This is the 7th development version of what will become
388 version 1.6 in the future. A development version not suitable for use in
389 production (unless you know exactly what you are doing). A stable version
390 will show as a 3-numbers version, such as "1.5.14-16f863", indicating the
391 14th level of fix on top of version 1.5. This is a production-ready version.
392
393 - the release date : 2015/10/08. It is represented in the universal
394 year/month/day format. Here this means August 8th, 2015. Given that stable
395 releases are issued every few months (1-2 months at the beginning, sometimes
396 6 months once the product becomes very stable), if you're seeing an old date
397 here, it means you're probably affected by a number of bugs or security
398 issues that have since been fixed and that it might be worth checking on the
399 official site.
400
401 - build options : they are relevant to people who build their packages
402 themselves, they can explain why things are not behaving as expected. For
403 example the development version above was built for Linux 2.6.28 or later,
Dan Lloyd8e48b872016-07-01 21:01:18 -0400404 targeting a generic CPU (no CPU-specific optimizations), and lacks any
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200405 code optimization (-O0) so it will perform poorly in terms of performance.
406
407 - libraries versions : zlib version is reported as found in the library
408 itself. In general zlib is considered a very stable product and upgrades
409 are almost never needed. OpenSSL reports two versions, the version used at
410 build time and the one being used, as found on the system. These ones may
411 differ by the last letter but never by the numbers. The build date is also
412 reported because most OpenSSL bugs are security issues and need to be taken
413 seriously, so this library absolutely needs to be kept up to date. Seeing a
414 4-months old version here is highly suspicious and indeed an update was
415 missed. PCRE provides very fast regular expressions and is highly
416 recommended. Certain of its extensions such as JIT are not present in all
417 versions and still young so some people prefer not to build with them,
Dan Lloyd8e48b872016-07-01 21:01:18 -0400418 which is why the build status is reported as well. Regarding the Lua
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200419 scripting language, HAProxy expects version 5.3 which is very young since
420 it was released a little time before HAProxy 1.6. It is important to check
421 on the Lua web site if some fixes are proposed for this branch.
422
423 - Available polling systems will affect the process's scalability when
424 dealing with more than about one thousand of concurrent connections. These
425 ones are only available when the correct system was indicated in the TARGET
426 variable during the build. The "epoll" mechanism is highly recommended on
427 Linux, and the kqueue mechanism is highly recommended on BSD. Lacking them
428 will result in poll() or even select() being used, causing a high CPU usage
429 when dealing with a lot of connections.
430
431
4324. Stopping and restarting HAProxy
433----------------------------------
434
435HAProxy supports a graceful and a hard stop. The hard stop is simple, when the
436SIGTERM signal is sent to the haproxy process, it immediately quits and all
437established connections are closed. The graceful stop is triggered when the
438SIGUSR1 signal is sent to the haproxy process. It consists in only unbinding
439from listening ports, but continue to process existing connections until they
440close. Once the last connection is closed, the process leaves.
441
442The hard stop method is used for the "stop" or "restart" actions of the service
443management script. The graceful stop is used for the "reload" action which
444tries to seamlessly reload a new configuration in a new process.
445
446Both of these signals may be sent by the new haproxy process itself during a
447reload or restart, so that they are sent at the latest possible moment and only
448if absolutely required. This is what is performed by the "-st" (hard) and "-sf"
449(graceful) options respectively.
450
William Lallemande202b1e2017-06-01 17:38:56 +0200451In master-worker mode, it is not needed to start a new haproxy process in
452order to reload the configuration. The master process reacts to the SIGUSR2
453signal by reexecuting itself with the -sf parameter followed by the PIDs of
454the workers. The master will then parse the configuration file and fork new
455workers.
456
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200457To understand better how these signals are used, it is important to understand
458the whole restart mechanism.
459
460First, an existing haproxy process is running. The administrator uses a system
461specific command such as "/etc/init.d/haproxy reload" to indicate he wants to
462take the new configuration file into effect. What happens then is the following.
463First, the service script (/etc/init.d/haproxy or equivalent) will verify that
464the configuration file parses correctly using "haproxy -c". After that it will
465try to start haproxy with this configuration file, using "-st" or "-sf".
466
467Then HAProxy tries to bind to all listening ports. If some fatal errors happen
468(eg: address not present on the system, permission denied), the process quits
469with an error. If a socket binding fails because a port is already in use, then
470the process will first send a SIGTTOU signal to all the pids specified in the
471"-st" or "-sf" pid list. This is what is called the "pause" signal. It instructs
472all existing haproxy processes to temporarily stop listening to their ports so
473that the new process can try to bind again. During this time, the old process
474continues to process existing connections. If the binding still fails (because
475for example a port is shared with another daemon), then the new process sends a
476SIGTTIN signal to the old processes to instruct them to resume operations just
477as if nothing happened. The old processes will then restart listening to the
478ports and continue to accept connections. Not that this mechanism is system
Dan Lloyd8e48b872016-07-01 21:01:18 -0400479dependent and some operating systems may not support it in multi-process mode.
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200480
481If the new process manages to bind correctly to all ports, then it sends either
482the SIGTERM (hard stop in case of "-st") or the SIGUSR1 (graceful stop in case
483of "-sf") to all processes to notify them that it is now in charge of operations
484and that the old processes will have to leave, either immediately or once they
485have finished their job.
486
487It is important to note that during this timeframe, there are two small windows
488of a few milliseconds each where it is possible that a few connection failures
489will be noticed during high loads. Typically observed failure rates are around
4901 failure during a reload operation every 10000 new connections per second,
491which means that a heavily loaded site running at 30000 new connections per
492second may see about 3 failed connection upon every reload. The two situations
493where this happens are :
494
495 - if the new process fails to bind due to the presence of the old process,
496 it will first have to go through the SIGTTOU+SIGTTIN sequence, which
497 typically lasts about one millisecond for a few tens of frontends, and
498 during which some ports will not be bound to the old process and not yet
499 bound to the new one. HAProxy works around this on systems that support the
500 SO_REUSEPORT socket options, as it allows the new process to bind without
501 first asking the old one to unbind. Most BSD systems have been supporting
502 this almost forever. Linux has been supporting this in version 2.0 and
503 dropped it around 2.2, but some patches were floating around by then. It
504 was reintroduced in kernel 3.9, so if you are observing a connection
Dan Lloyd8e48b872016-07-01 21:01:18 -0400505 failure rate above the one mentioned above, please ensure that your kernel
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200506 is 3.9 or newer, or that relevant patches were backported to your kernel
507 (less likely).
508
509 - when the old processes close the listening ports, the kernel may not always
510 redistribute any pending connection that was remaining in the socket's
511 backlog. Under high loads, a SYN packet may happen just before the socket
512 is closed, and will lead to an RST packet being sent to the client. In some
513 critical environments where even one drop is not acceptable, these ones are
514 sometimes dealt with using firewall rules to block SYN packets during the
515 reload, forcing the client to retransmit. This is totally system-dependent,
516 as some systems might be able to visit other listening queues and avoid
517 this RST. A second case concerns the ACK from the client on a local socket
518 that was in SYN_RECV state just before the close. This ACK will lead to an
519 RST packet while the haproxy process is still not aware of it. This one is
Dan Lloyd8e48b872016-07-01 21:01:18 -0400520 harder to get rid of, though the firewall filtering rules mentioned above
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200521 will work well if applied one second or so before restarting the process.
522
523For the vast majority of users, such drops will never ever happen since they
524don't have enough load to trigger the race conditions. And for most high traffic
525users, the failure rate is still fairly within the noise margin provided that at
526least SO_REUSEPORT is properly supported on their systems.
527
528
5295. File-descriptor limitations
530------------------------------
531
532In order to ensure that all incoming connections will successfully be served,
533HAProxy computes at load time the total number of file descriptors that will be
534needed during the process's life. A regular Unix process is generally granted
5351024 file descriptors by default, and a privileged process can raise this limit
536itself. This is one reason for starting HAProxy as root and letting it adjust
537the limit. The default limit of 1024 file descriptors roughly allow about 500
538concurrent connections to be processed. The computation is based on the global
539maxconn parameter which limits the total number of connections per process, the
540number of listeners, the number of servers which have a health check enabled,
541the agent checks, the peers, the loggers and possibly a few other technical
542requirements. A simple rough estimate of this number consists in simply
543doubling the maxconn value and adding a few tens to get the approximate number
544of file descriptors needed.
545
546Originally HAProxy did not know how to compute this value, and it was necessary
547to pass the value using the "ulimit-n" setting in the global section. This
548explains why even today a lot of configurations are seen with this setting
549present. Unfortunately it was often miscalculated resulting in connection
550failures when approaching maxconn instead of throttling incoming connection
551while waiting for the needed resources. For this reason it is important to
Dan Lloyd8e48b872016-07-01 21:01:18 -0400552remove any vestigial "ulimit-n" setting that can remain from very old versions.
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200553
554Raising the number of file descriptors to accept even moderate loads is
555mandatory but comes with some OS-specific adjustments. First, the select()
556polling system is limited to 1024 file descriptors. In fact on Linux it used
557to be capable of handling more but since certain OS ship with excessively
558restrictive SELinux policies forbidding the use of select() with more than
5591024 file descriptors, HAProxy now refuses to start in this case in order to
560avoid any issue at run time. On all supported operating systems, poll() is
561available and will not suffer from this limitation. It is automatically picked
Dan Lloyd8e48b872016-07-01 21:01:18 -0400562so there is nothing to do to get a working configuration. But poll's becomes
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200563very slow when the number of file descriptors increases. While HAProxy does its
564best to limit this performance impact (eg: via the use of the internal file
565descriptor cache and batched processing), a good rule of thumb is that using
566poll() with more than a thousand concurrent connections will use a lot of CPU.
567
568For Linux systems base on kernels 2.6 and above, the epoll() system call will
569be used. It's a much more scalable mechanism relying on callbacks in the kernel
570that guarantee a constant wake up time regardless of the number of registered
571monitored file descriptors. It is automatically used where detected, provided
572that HAProxy had been built for one of the Linux flavors. Its presence and
573support can be verified using "haproxy -vv".
574
575For BSD systems which support it, kqueue() is available as an alternative. It
576is much faster than poll() and even slightly faster than epoll() thanks to its
577batched handling of changes. At least FreeBSD and OpenBSD support it. Just like
578with Linux's epoll(), its support and availability are reported in the output
579of "haproxy -vv".
580
581Having a good poller is one thing, but it is mandatory that the process can
582reach the limits. When HAProxy starts, it immediately sets the new process's
583file descriptor limits and verifies if it succeeds. In case of failure, it
584reports it before forking so that the administrator can see the problem. As
585long as the process is started by as root, there should be no reason for this
586setting to fail. However, it can fail if the process is started by an
587unprivileged user. If there is a compelling reason for *not* starting haproxy
588as root (eg: started by end users, or by a per-application account), then the
589file descriptor limit can be raised by the system administrator for this
590specific user. The effectiveness of the setting can be verified by issuing
591"ulimit -n" from the user's command line. It should reflect the new limit.
592
593Warning: when an unprivileged user's limits are changed in this user's account,
594it is fairly common that these values are only considered when the user logs in
595and not at all in some scripts run at system boot time nor in crontabs. This is
596totally dependent on the operating system, keep in mind to check "ulimit -n"
597before starting haproxy when running this way. The general advice is never to
598start haproxy as an unprivileged user for production purposes. Another good
599reason is that it prevents haproxy from enabling some security protections.
600
601Once it is certain that the system will allow the haproxy process to use the
602requested number of file descriptors, two new system-specific limits may be
603encountered. The first one is the system-wide file descriptor limit, which is
604the total number of file descriptors opened on the system, covering all
605processes. When this limit is reached, accept() or socket() will typically
606return ENFILE. The second one is the per-process hard limit on the number of
607file descriptors, it prevents setrlimit() from being set higher. Both are very
608dependent on the operating system. On Linux, the system limit is set at boot
609based on the amount of memory. It can be changed with the "fs.file-max" sysctl.
610And the per-process hard limit is set to 1048576 by default, but it can be
611changed using the "fs.nr_open" sysctl.
612
613File descriptor limitations may be observed on a running process when they are
614set too low. The strace utility will report that accept() and socket() return
615"-1 EMFILE" when the process's limits have been reached. In this case, simply
616raising the "ulimit-n" value (or removing it) will solve the problem. If these
617system calls return "-1 ENFILE" then it means that the kernel's limits have
618been reached and that something must be done on a system-wide parameter. These
619trouble must absolutely be addressed, as they result in high CPU usage (when
620accept() fails) and failed connections that are generally visible to the user.
621One solution also consists in lowering the global maxconn value to enforce
622serialization, and possibly to disable HTTP keep-alive to force connections
623to be released and reused faster.
624
625
6266. Memory management
627--------------------
628
629HAProxy uses a simple and fast pool-based memory management. Since it relies on
630a small number of different object types, it's much more efficient to pick new
631objects from a pool which already contains objects of the appropriate size than
632to call malloc() for each different size. The pools are organized as a stack or
633LIFO, so that newly allocated objects are taken from recently released objects
634still hot in the CPU caches. Pools of similar sizes are merged together, in
635order to limit memory fragmentation.
636
637By default, since the focus is set on performance, each released object is put
638back into the pool it came from, and allocated objects are never freed since
639they are expected to be reused very soon.
640
641On the CLI, it is possible to check how memory is being used in pools thanks to
642the "show pools" command :
643
644 > show pools
645 Dumping pools usage. Use SIGQUIT to flush them.
Willy Tarreau0a93b642018-10-16 07:58:39 +0200646 - Pool cache_st (16 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 1 users, @0x9ccc40=03 [SHARED]
647 - Pool pipe (32 bytes) : 5 allocated (160 bytes), 5 used, 0 failures, 2 users, @0x9ccac0=00 [SHARED]
648 - Pool comp_state (48 bytes) : 3 allocated (144 bytes), 3 used, 0 failures, 5 users, @0x9cccc0=04 [SHARED]
649 - Pool filter (64 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 3 users, @0x9ccbc0=02 [SHARED]
650 - Pool vars (80 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 2 users, @0x9ccb40=01 [SHARED]
651 - Pool uniqueid (128 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 2 users, @0x9cd240=15 [SHARED]
652 - Pool task (144 bytes) : 55 allocated (7920 bytes), 55 used, 0 failures, 1 users, @0x9cd040=11 [SHARED]
653 - Pool session (160 bytes) : 1 allocated (160 bytes), 1 used, 0 failures, 1 users, @0x9cd140=13 [SHARED]
654 - Pool h2s (208 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 2 users, @0x9ccec0=08 [SHARED]
655 - Pool h2c (288 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 1 users, @0x9cce40=07 [SHARED]
656 - Pool spoe_ctx (304 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 2 users, @0x9ccf40=09 [SHARED]
657 - Pool connection (400 bytes) : 2 allocated (800 bytes), 2 used, 0 failures, 1 users, @0x9cd1c0=14 [SHARED]
658 - Pool hdr_idx (416 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 1 users, @0x9cd340=17 [SHARED]
659 - Pool dns_resolut (480 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 1 users, @0x9ccdc0=06 [SHARED]
660 - Pool dns_answer_ (576 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 1 users, @0x9ccd40=05 [SHARED]
661 - Pool stream (960 bytes) : 1 allocated (960 bytes), 1 used, 0 failures, 1 users, @0x9cd0c0=12 [SHARED]
662 - Pool requri (1024 bytes) : 0 allocated (0 bytes), 0 used, 0 failures, 1 users, @0x9cd2c0=16 [SHARED]
663 - Pool buffer (8030 bytes) : 3 allocated (24090 bytes), 2 used, 0 failures, 1 users, @0x9cd3c0=18 [SHARED]
664 - Pool trash (8062 bytes) : 1 allocated (8062 bytes), 1 used, 0 failures, 1 users, @0x9cd440=19
665 Total: 19 pools, 42296 bytes allocated, 34266 used.
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200666
667The pool name is only indicative, it's the name of the first object type using
668this pool. The size in parenthesis is the object size for objects in this pool.
669Object sizes are always rounded up to the closest multiple of 16 bytes. The
670number of objects currently allocated and the equivalent number of bytes is
671reported so that it is easy to know which pool is responsible for the highest
672memory usage. The number of objects currently in use is reported as well in the
673"used" field. The difference between "allocated" and "used" corresponds to the
Willy Tarreau0a93b642018-10-16 07:58:39 +0200674objects that have been freed and are available for immediate use. The address
675at the end of the line is the pool's address, and the following number is the
676pool index when it exists, or is reported as -1 if no index was assigned.
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200677
678It is possible to limit the amount of memory allocated per process using the
679"-m" command line option, followed by a number of megabytes. It covers all of
680the process's addressable space, so that includes memory used by some libraries
681as well as the stack, but it is a reliable limit when building a resource
682constrained system. It works the same way as "ulimit -v" on systems which have
683it, or "ulimit -d" for the other ones.
684
685If a memory allocation fails due to the memory limit being reached or because
686the system doesn't have any enough memory, then haproxy will first start to
687free all available objects from all pools before attempting to allocate memory
688again. This mechanism of releasing unused memory can be triggered by sending
689the signal SIGQUIT to the haproxy process. When doing so, the pools state prior
690to the flush will also be reported to stderr when the process runs in
691foreground.
692
693During a reload operation, the process switched to the graceful stop state also
694automatically performs some flushes after releasing any connection so that all
695possible memory is released to save it for the new process.
696
697
6987. CPU usage
699------------
700
701HAProxy normally spends most of its time in the system and a smaller part in
702userland. A finely tuned 3.5 GHz CPU can sustain a rate about 80000 end-to-end
703connection setups and closes per second at 100% CPU on a single core. When one
704core is saturated, typical figures are :
705 - 95% system, 5% user for long TCP connections or large HTTP objects
706 - 85% system and 15% user for short TCP connections or small HTTP objects in
707 close mode
708 - 70% system and 30% user for small HTTP objects in keep-alive mode
709
710The amount of rules processing and regular expressions will increase the user
711land part. The presence of firewall rules, connection tracking, complex routing
712tables in the system will instead increase the system part.
713
714On most systems, the CPU time observed during network transfers can be cut in 4
715parts :
716 - the interrupt part, which concerns all the processing performed upon I/O
717 receipt, before the target process is even known. Typically Rx packets are
718 accounted for in interrupt. On some systems such as Linux where interrupt
719 processing may be deferred to a dedicated thread, it can appear as softirq,
720 and the thread is called ksoftirqd/0 (for CPU 0). The CPU taking care of
721 this load is generally defined by the hardware settings, though in the case
722 of softirq it is often possible to remap the processing to another CPU.
723 This interrupt part will often be perceived as parasitic since it's not
724 associated with any process, but it actually is some processing being done
725 to prepare the work for the process.
726
727 - the system part, which concerns all the processing done using kernel code
728 called from userland. System calls are accounted as system for example. All
729 synchronously delivered Tx packets will be accounted for as system time. If
730 some packets have to be deferred due to queues filling up, they may then be
731 processed in interrupt context later (eg: upon receipt of an ACK opening a
732 TCP window).
733
734 - the user part, which exclusively runs application code in userland. HAProxy
735 runs exclusively in this part, though it makes heavy use of system calls.
736 Rules processing, regular expressions, compression, encryption all add to
737 the user portion of CPU consumption.
738
739 - the idle part, which is what the CPU does when there is nothing to do. For
740 example HAProxy waits for an incoming connection, or waits for some data to
741 leave, meaning the system is waiting for an ACK from the client to push
742 these data.
743
744In practice regarding HAProxy's activity, it is in general reasonably accurate
745(but totally inexact) to consider that interrupt/softirq are caused by Rx
746processing in kernel drivers, that user-land is caused by layer 7 processing
747in HAProxy, and that system time is caused by network processing on the Tx
748path.
749
750Since HAProxy runs around an event loop, it waits for new events using poll()
751(or any alternative) and processes all these events as fast as possible before
752going back to poll() waiting for new events. It measures the time spent waiting
753in poll() compared to the time spent doing processing events. The ratio of
754polling time vs total time is called the "idle" time, it's the amount of time
755spent waiting for something to happen. This ratio is reported in the stats page
756on the "idle" line, or "Idle_pct" on the CLI. When it's close to 100%, it means
757the load is extremely low. When it's close to 0%, it means that there is
758constantly some activity. While it cannot be very accurate on an overloaded
759system due to other processes possibly preempting the CPU from the haproxy
760process, it still provides a good estimate about how HAProxy considers it is
761working : if the load is low and the idle ratio is low as well, it may indicate
762that HAProxy has a lot of work to do, possibly due to very expensive rules that
763have to be processed. Conversely, if HAProxy indicates the idle is close to
764100% while things are slow, it means that it cannot do anything to speed things
765up because it is already waiting for incoming data to process. In the example
766below, haproxy is completely idle :
767
768 $ echo "show info" | socat - /var/run/haproxy.sock | grep ^Idle
769 Idle_pct: 100
770
771When the idle ratio starts to become very low, it is important to tune the
772system and place processes and interrupts correctly to save the most possible
773CPU resources for all tasks. If a firewall is present, it may be worth trying
774to disable it or to tune it to ensure it is not responsible for a large part
775of the performance limitation. It's worth noting that unloading a stateful
776firewall generally reduces both the amount of interrupt/softirq and of system
777usage since such firewalls act both on the Rx and the Tx paths. On Linux,
778unloading the nf_conntrack and ip_conntrack modules will show whether there is
779anything to gain. If so, then the module runs with default settings and you'll
780have to figure how to tune it for better performance. In general this consists
781in considerably increasing the hash table size. On FreeBSD, "pfctl -d" will
782disable the "pf" firewall and its stateful engine at the same time.
783
784If it is observed that a lot of time is spent in interrupt/softirq, it is
785important to ensure that they don't run on the same CPU. Most systems tend to
786pin the tasks on the CPU where they receive the network traffic because for
787certain workloads it improves things. But with heavily network-bound workloads
788it is the opposite as the haproxy process will have to fight against its kernel
789counterpart. Pinning haproxy to one CPU core and the interrupts to another one,
790all sharing the same L3 cache tends to sensibly increase network performance
791because in practice the amount of work for haproxy and the network stack are
792quite close, so they can almost fill an entire CPU each. On Linux this is done
793using taskset (for haproxy) or using cpu-map (from the haproxy config), and the
794interrupts are assigned under /proc/irq. Many network interfaces support
795multiple queues and multiple interrupts. In general it helps to spread them
796across a small number of CPU cores provided they all share the same L3 cache.
797Please always stop irq_balance which always does the worst possible thing on
798such workloads.
799
800For CPU-bound workloads consisting in a lot of SSL traffic or a lot of
801compression, it may be worth using multiple processes dedicated to certain
802tasks, though there is no universal rule here and experimentation will have to
803be performed.
804
805In order to increase the CPU capacity, it is possible to make HAProxy run as
806several processes, using the "nbproc" directive in the global section. There
807are some limitations though :
808 - health checks are run per process, so the target servers will get as many
809 checks as there are running processes ;
810 - maxconn values and queues are per-process so the correct value must be set
811 to avoid overloading the servers ;
812 - outgoing connections should avoid using port ranges to avoid conflicts
813 - stick-tables are per process and are not shared between processes ;
814 - each peers section may only run on a single process at a time ;
815 - the CLI operations will only act on a single process at a time.
816
817With this in mind, it appears that the easiest setup often consists in having
818one first layer running on multiple processes and in charge for the heavy
819processing, passing the traffic to a second layer running in a single process.
820This mechanism is suited to SSL and compression which are the two CPU-heavy
821features. Instances can easily be chained over UNIX sockets (which are cheaper
fengpeiyuancc123c62016-01-15 16:40:53 +0800822than TCP sockets and which do not waste ports), and the proxy protocol which is
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200823useful to pass client information to the next stage. When doing so, it is
824generally a good idea to bind all the single-process tasks to process number 1
825and extra tasks to next processes, as this will make it easier to generate
826similar configurations for different machines.
827
828On Linux versions 3.9 and above, running HAProxy in multi-process mode is much
829more efficient when each process uses a distinct listening socket on the same
830IP:port ; this will make the kernel evenly distribute the load across all
831processes instead of waking them all up. Please check the "process" option of
832the "bind" keyword lines in the configuration manual for more information.
833
834
8358. Logging
836----------
837
838For logging, HAProxy always relies on a syslog server since it does not perform
839any file-system access. The standard way of using it is to send logs over UDP
840to the log server (by default on port 514). Very commonly this is configured to
841127.0.0.1 where the local syslog daemon is running, but it's also used over the
842network to log to a central server. The central server provides additional
843benefits especially in active-active scenarios where it is desirable to keep
844the logs merged in arrival order. HAProxy may also make use of a UNIX socket to
845send its logs to the local syslog daemon, but it is not recommended at all,
846because if the syslog server is restarted while haproxy runs, the socket will
847be replaced and new logs will be lost. Since HAProxy will be isolated inside a
848chroot jail, it will not have the ability to reconnect to the new socket. It
849has also been observed in field that the log buffers in use on UNIX sockets are
850very small and lead to lost messages even at very light loads. But this can be
851fine for testing however.
852
853It is recommended to add the following directive to the "global" section to
854make HAProxy log to the local daemon using facility "local0" :
855
856 log 127.0.0.1:514 local0
857
858and then to add the following one to each "defaults" section or to each frontend
859and backend section :
860
861 log global
862
863This way, all logs will be centralized through the global definition of where
864the log server is.
865
866Some syslog daemons do not listen to UDP traffic by default, so depending on
867the daemon being used, the syntax to enable this will vary :
868
869 - on sysklogd, you need to pass argument "-r" on the daemon's command line
870 so that it listens to a UDP socket for "remote" logs ; note that there is
871 no way to limit it to address 127.0.0.1 so it will also receive logs from
872 remote systems ;
873
874 - on rsyslogd, the following lines must be added to the configuration file :
875
876 $ModLoad imudp
877 $UDPServerAddress *
878 $UDPServerRun 514
879
880 - on syslog-ng, a new source can be created the following way, it then needs
881 to be added as a valid source in one of the "log" directives :
882
883 source s_udp {
884 udp(ip(127.0.0.1) port(514));
885 };
886
887Please consult your syslog daemon's manual for more information. If no logs are
888seen in the system's log files, please consider the following tests :
889
890 - restart haproxy. Each frontend and backend logs one line indicating it's
891 starting. If these logs are received, it means logs are working.
892
893 - run "strace -tt -s100 -etrace=sendmsg -p <haproxy's pid>" and perform some
894 activity that you expect to be logged. You should see the log messages
895 being sent using sendmsg() there. If they don't appear, restart using
896 strace on top of haproxy. If you still see no logs, it definitely means
897 that something is wrong in your configuration.
898
899 - run tcpdump to watch for port 514, for example on the loopback interface if
900 the traffic is being sent locally : "tcpdump -As0 -ni lo port 514". If the
901 packets are seen there, it's the proof they're sent then the syslogd daemon
902 needs to be troubleshooted.
903
904While traffic logs are sent from the frontends (where the incoming connections
905are accepted), backends also need to be able to send logs in order to report a
906server state change consecutive to a health check. Please consult HAProxy's
907configuration manual for more information regarding all possible log settings.
908
Dan Lloyd8e48b872016-07-01 21:01:18 -0400909It is convenient to chose a facility that is not used by other daemons. HAProxy
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200910examples often suggest "local0" for traffic logs and "local1" for admin logs
911because they're never seen in field. A single facility would be enough as well.
912Having separate logs is convenient for log analysis, but it's also important to
913remember that logs may sometimes convey confidential information, and as such
Dan Lloyd8e48b872016-07-01 21:01:18 -0400914they must not be mixed with other logs that may accidentally be handed out to
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200915unauthorized people.
916
917For in-field troubleshooting without impacting the server's capacity too much,
918it is recommended to make use of the "halog" utility provided with HAProxy.
919This is sort of a grep-like utility designed to process HAProxy log files at
920a very fast data rate. Typical figures range between 1 and 2 GB of logs per
921second. It is capable of extracting only certain logs (eg: search for some
922classes of HTTP status codes, connection termination status, search by response
923time ranges, look for errors only), count lines, limit the output to a number
924of lines, and perform some more advanced statistics such as sorting servers
925by response time or error counts, sorting URLs by time or count, sorting client
926addresses by access count, and so on. It is pretty convenient to quickly spot
927anomalies such as a bot looping on the site, and block them.
928
929
9309. Statistics and monitoring
931----------------------------
932
Willy Tarreau44aed902015-10-13 14:45:29 +0200933It is possible to query HAProxy about its status. The most commonly used
934mechanism is the HTTP statistics page. This page also exposes an alternative
935CSV output format for monitoring tools. The same format is provided on the
936Unix socket.
937
938
9399.1. CSV format
940---------------
941
942The statistics may be consulted either from the unix socket or from the HTTP
943page. Both means provide a CSV format whose fields follow. The first line
944begins with a sharp ('#') and has one word per comma-delimited field which
945represents the title of the column. All other lines starting at the second one
946use a classical CSV format using a comma as the delimiter, and the double quote
947('"') as an optional text delimiter, but only if the enclosed text is ambiguous
948(if it contains a quote or a comma). The double-quote character ('"') in the
949text is doubled ('""'), which is the format that most tools recognize. Please
950do not insert any column before these ones in order not to break tools which
951use hard-coded column positions.
952
953In brackets after each field name are the types which may have a value for
954that field. The types are L (Listeners), F (Frontends), B (Backends), and
955S (Servers).
956
957 0. pxname [LFBS]: proxy name
958 1. svname [LFBS]: service name (FRONTEND for frontend, BACKEND for backend,
959 any name for server/listener)
960 2. qcur [..BS]: current queued requests. For the backend this reports the
961 number queued without a server assigned.
962 3. qmax [..BS]: max value of qcur
963 4. scur [LFBS]: current sessions
964 5. smax [LFBS]: max sessions
965 6. slim [LFBS]: configured session limit
Willy Tarreauc73810f2016-01-11 13:52:04 +0100966 7. stot [LFBS]: cumulative number of sessions
Willy Tarreau44aed902015-10-13 14:45:29 +0200967 8. bin [LFBS]: bytes in
968 9. bout [LFBS]: bytes out
969 10. dreq [LFB.]: requests denied because of security concerns.
970 - For tcp this is because of a matched tcp-request content rule.
971 - For http this is because of a matched http-request or tarpit rule.
972 11. dresp [LFBS]: responses denied because of security concerns.
973 - For http this is because of a matched http-request rule, or
974 "option checkcache".
975 12. ereq [LF..]: request errors. Some of the possible causes are:
976 - early termination from the client, before the request has been sent.
977 - read error from the client
978 - client timeout
979 - client closed connection
980 - various bad requests from the client.
981 - request was tarpitted.
982 13. econ [..BS]: number of requests that encountered an error trying to
983 connect to a backend server. The backend stat is the sum of the stat
984 for all servers of that backend, plus any connection errors not
985 associated with a particular server (such as the backend having no
986 active servers).
987 14. eresp [..BS]: response errors. srv_abrt will be counted here also.
988 Some other errors are:
989 - write error on the client socket (won't be counted for the server stat)
990 - failure applying filters to the response.
991 15. wretr [..BS]: number of times a connection to a server was retried.
992 16. wredis [..BS]: number of times a request was redispatched to another
993 server. The server value counts the number of times that server was
994 switched away from.
Willy Tarreaub96dd282016-11-09 14:45:51 +0100995 17. status [LFBS]: status (UP/DOWN/NOLB/MAINT/MAINT(via)/MAINT(resolution)...)
Willy Tarreau44aed902015-10-13 14:45:29 +0200996 18. weight [..BS]: total weight (backend), server weight (server)
997 19. act [..BS]: number of active servers (backend), server is active (server)
998 20. bck [..BS]: number of backup servers (backend), server is backup (server)
999 21. chkfail [...S]: number of failed checks. (Only counts checks failed when
1000 the server is up.)
1001 22. chkdown [..BS]: number of UP->DOWN transitions. The backend counter counts
1002 transitions to the whole backend being down, rather than the sum of the
1003 counters for each server.
1004 23. lastchg [..BS]: number of seconds since the last UP<->DOWN transition
1005 24. downtime [..BS]: total downtime (in seconds). The value for the backend
1006 is the downtime for the whole backend, not the sum of the server downtime.
1007 25. qlimit [...S]: configured maxqueue for the server, or nothing in the
1008 value is 0 (default, meaning no limit)
1009 26. pid [LFBS]: process id (0 for first instance, 1 for second, ...)
1010 27. iid [LFBS]: unique proxy id
1011 28. sid [L..S]: server id (unique inside a proxy)
1012 29. throttle [...S]: current throttle percentage for the server, when
1013 slowstart is active, or no value if not in slowstart.
1014 30. lbtot [..BS]: total number of times a server was selected, either for new
1015 sessions, or when re-dispatching. The server counter is the number
1016 of times that server was selected.
1017 31. tracked [...S]: id of proxy/server if tracking is enabled.
1018 32. type [LFBS]: (0=frontend, 1=backend, 2=server, 3=socket/listener)
1019 33. rate [.FBS]: number of sessions per second over last elapsed second
1020 34. rate_lim [.F..]: configured limit on new sessions per second
1021 35. rate_max [.FBS]: max number of new sessions per second
1022 36. check_status [...S]: status of last health check, one of:
1023 UNK -> unknown
1024 INI -> initializing
1025 SOCKERR -> socket error
1026 L4OK -> check passed on layer 4, no upper layers testing enabled
1027 L4TOUT -> layer 1-4 timeout
1028 L4CON -> layer 1-4 connection problem, for example
1029 "Connection refused" (tcp rst) or "No route to host" (icmp)
1030 L6OK -> check passed on layer 6
1031 L6TOUT -> layer 6 (SSL) timeout
1032 L6RSP -> layer 6 invalid response - protocol error
1033 L7OK -> check passed on layer 7
1034 L7OKC -> check conditionally passed on layer 7, for example 404 with
1035 disable-on-404
1036 L7TOUT -> layer 7 (HTTP/SMTP) timeout
1037 L7RSP -> layer 7 invalid response - protocol error
1038 L7STS -> layer 7 response error, for example HTTP 5xx
Daniel Schnellerb6c8b0d2017-09-01 19:13:55 +02001039 Notice: If a check is currently running, the last known status will be
1040 reported, prefixed with "* ". e. g. "* L7OK".
Willy Tarreau44aed902015-10-13 14:45:29 +02001041 37. check_code [...S]: layer5-7 code, if available
1042 38. check_duration [...S]: time in ms took to finish last health check
1043 39. hrsp_1xx [.FBS]: http responses with 1xx code
1044 40. hrsp_2xx [.FBS]: http responses with 2xx code
1045 41. hrsp_3xx [.FBS]: http responses with 3xx code
1046 42. hrsp_4xx [.FBS]: http responses with 4xx code
1047 43. hrsp_5xx [.FBS]: http responses with 5xx code
1048 44. hrsp_other [.FBS]: http responses with other codes (protocol error)
1049 45. hanafail [...S]: failed health checks details
1050 46. req_rate [.F..]: HTTP requests per second over last elapsed second
1051 47. req_rate_max [.F..]: max number of HTTP requests per second observed
Willy Tarreaufb981bd2016-12-12 14:31:46 +01001052 48. req_tot [.FB.]: total number of HTTP requests received
Willy Tarreau44aed902015-10-13 14:45:29 +02001053 49. cli_abrt [..BS]: number of data transfers aborted by the client
1054 50. srv_abrt [..BS]: number of data transfers aborted by the server
1055 (inc. in eresp)
1056 51. comp_in [.FB.]: number of HTTP response bytes fed to the compressor
1057 52. comp_out [.FB.]: number of HTTP response bytes emitted by the compressor
1058 53. comp_byp [.FB.]: number of bytes that bypassed the HTTP compressor
1059 (CPU/BW limit)
1060 54. comp_rsp [.FB.]: number of HTTP responses that were compressed
1061 55. lastsess [..BS]: number of seconds since last session assigned to
1062 server/backend
1063 56. last_chk [...S]: last health check contents or textual error
1064 57. last_agt [...S]: last agent check contents or textual error
1065 58. qtime [..BS]: the average queue time in ms over the 1024 last requests
1066 59. ctime [..BS]: the average connect time in ms over the 1024 last requests
1067 60. rtime [..BS]: the average response time in ms over the 1024 last requests
1068 (0 for TCP)
1069 61. ttime [..BS]: the average total session time in ms over the 1024 last
1070 requests
Willy Tarreau7f618842016-01-08 11:40:03 +01001071 62. agent_status [...S]: status of last agent check, one of:
1072 UNK -> unknown
1073 INI -> initializing
1074 SOCKERR -> socket error
1075 L4OK -> check passed on layer 4, no upper layers testing enabled
1076 L4TOUT -> layer 1-4 timeout
1077 L4CON -> layer 1-4 connection problem, for example
1078 "Connection refused" (tcp rst) or "No route to host" (icmp)
1079 L7OK -> agent reported "up"
1080 L7STS -> agent reported "fail", "stop", or "down"
1081 63. agent_code [...S]: numeric code reported by agent if any (unused for now)
1082 64. agent_duration [...S]: time in ms taken to finish last check
Willy Tarreaudd7354b2016-01-08 13:47:26 +01001083 65. check_desc [...S]: short human-readable description of check_status
1084 66. agent_desc [...S]: short human-readable description of agent_status
Willy Tarreau3141f592016-01-08 14:25:28 +01001085 67. check_rise [...S]: server's "rise" parameter used by checks
1086 68. check_fall [...S]: server's "fall" parameter used by checks
1087 69. check_health [...S]: server's health check value between 0 and rise+fall-1
1088 70. agent_rise [...S]: agent's "rise" parameter, normally 1
1089 71. agent_fall [...S]: agent's "fall" parameter, normally 1
1090 72. agent_health [...S]: agent's health parameter, between 0 and rise+fall-1
Willy Tarreaua6f5a732016-01-08 16:59:56 +01001091 73. addr [L..S]: address:port or "unix". IPv6 has brackets around the address.
Willy Tarreaue4847c62016-01-08 15:43:54 +01001092 74: cookie [..BS]: server's cookie value or backend's cookie name
Willy Tarreauf8211df2016-01-11 14:09:38 +01001093 75: mode [LFBS]: proxy mode (tcp, http, health, unknown)
Willy Tarreauf1516d92016-01-11 14:48:36 +01001094 76: algo [..B.]: load balancing algorithm
Willy Tarreauc73810f2016-01-11 13:52:04 +01001095 77: conn_rate [.F..]: number of connections over the last elapsed second
1096 78: conn_rate_max [.F..]: highest known conn_rate
1097 79: conn_tot [.F..]: cumulative number of connections
Willy Tarreau5b9bdff2016-01-11 14:40:47 +01001098 80: intercepted [.FB.]: cum. number of intercepted requests (monitor, stats)
Willy Tarreau8a90b8e2016-10-21 18:15:32 +02001099 81: dcon [LF..]: requests denied by "tcp-request connection" rules
Willy Tarreaua5bc36b2016-10-21 18:16:27 +02001100 82: dses [LF..]: requests denied by "tcp-request session" rules
Willy Tarreauea96a822018-05-28 15:15:43 +02001101 83: wrew [LFBS]: cumulative number of failed header rewriting warnings
Willy Tarreau44aed902015-10-13 14:45:29 +02001102
1103
Willy Tarreau5d8b9792016-03-11 11:09:34 +010011049.2) Typed output format
1105------------------------
1106
1107Both "show info" and "show stat" support a mode where each output value comes
1108with its type and sufficient information to know how the value is supposed to
1109be aggregated between processes and how it evolves.
1110
1111In all cases, the output consists in having a single value per line with all
1112the information split into fields delimited by colons (':').
1113
1114The first column designates the object or metric being dumped. Its format is
1115specific to the command producing this output and will not be described in this
1116section. Usually it will consist in a series of identifiers and field names.
1117
1118The second column contains 3 characters respectively indicating the origin, the
1119nature and the scope of the value being reported. The first character (the
1120origin) indicates where the value was extracted from. Possible characters are :
1121
1122 M The value is a metric. It is valid at one instant any may change depending
1123 on its nature .
1124
1125 S The value is a status. It represents a discrete value which by definition
1126 cannot be aggregated. It may be the status of a server ("UP" or "DOWN"),
1127 the PID of the process, etc.
1128
1129 K The value is a sorting key. It represents an identifier which may be used
1130 to group some values together because it is unique among its class. All
1131 internal identifiers are keys. Some names can be listed as keys if they
1132 are unique (eg: a frontend name is unique). In general keys come from the
Dan Lloyd8e48b872016-07-01 21:01:18 -04001133 configuration, even though some of them may automatically be assigned. For
Willy Tarreau5d8b9792016-03-11 11:09:34 +01001134 most purposes keys may be considered as equivalent to configuration.
1135
1136 C The value comes from the configuration. Certain configuration values make
1137 sense on the output, for example a concurrent connection limit or a cookie
1138 name. By definition these values are the same in all processes started
1139 from the same configuration file.
1140
1141 P The value comes from the product itself. There are very few such values,
1142 most common use is to report the product name, version and release date.
1143 These elements are also the same between all processes.
1144
1145The second character (the nature) indicates the nature of the information
1146carried by the field in order to let an aggregator decide on what operation to
1147use to aggregate multiple values. Possible characters are :
1148
1149 A The value represents an age since a last event. This is a bit different
1150 from the duration in that an age is automatically computed based on the
1151 current date. A typical example is how long ago did the last session
1152 happen on a server. Ages are generally aggregated by taking the minimum
1153 value and do not need to be stored.
1154
1155 a The value represents an already averaged value. The average response times
1156 and server weights are of this nature. Averages can typically be averaged
1157 between processes.
1158
1159 C The value represents a cumulative counter. Such measures perpetually
1160 increase until they wrap around. Some monitoring protocols need to tell
1161 the difference between a counter and a gauge to report a different type.
1162 In general counters may simply be summed since they represent events or
1163 volumes. Examples of metrics of this nature are connection counts or byte
1164 counts.
1165
1166 D The value represents a duration for a status. There are a few usages of
1167 this, most of them include the time taken by the last health check and
1168 the time a server has spent down. Durations are generally not summed,
1169 most of the time the maximum will be retained to compute an SLA.
1170
1171 G The value represents a gauge. It's a measure at one instant. The memory
1172 usage or the current number of active connections are of this nature.
1173 Metrics of this type are typically summed during aggregation.
1174
1175 L The value represents a limit (generally a configured one). By nature,
1176 limits are harder to aggregate since they are specific to the point where
1177 they were retrieved. In certain situations they may be summed or be kept
1178 separate.
1179
1180 M The value represents a maximum. In general it will apply to a gauge and
1181 keep the highest known value. An example of such a metric could be the
1182 maximum amount of concurrent connections that was encountered in the
1183 product's life time. To correctly aggregate maxima, you are supposed to
1184 output a range going from the maximum of all maxima and the sum of all
1185 of them. There is indeed no way to know if they were encountered
1186 simultaneously or not.
1187
1188 m The value represents a minimum. In general it will apply to a gauge and
1189 keep the lowest known value. An example of such a metric could be the
1190 minimum amount of free memory pools that was encountered in the product's
1191 life time. To correctly aggregate minima, you are supposed to output a
1192 range going from the minimum of all minima and the sum of all of them.
1193 There is indeed no way to know if they were encountered simultaneously
1194 or not.
1195
1196 N The value represents a name, so it is a string. It is used to report
1197 proxy names, server names and cookie names. Names have configuration or
1198 keys as their origin and are supposed to be the same among all processes.
1199
1200 O The value represents a free text output. Outputs from various commands,
1201 returns from health checks, node descriptions are of such nature.
1202
1203 R The value represents an event rate. It's a measure at one instant. It is
1204 quite similar to a gauge except that the recipient knows that this measure
1205 moves slowly and may decide not to keep all values. An example of such a
1206 metric is the measured amount of connections per second. Metrics of this
1207 type are typically summed during aggregation.
1208
1209 T The value represents a date or time. A field emitting the current date
1210 would be of this type. The method to aggregate such information is left
1211 as an implementation choice. For now no field uses this type.
1212
1213The third character (the scope) indicates what extent the value reflects. Some
1214elements may be per process while others may be per configuration or per system.
1215The distinction is important to know whether or not a single value should be
1216kept during aggregation or if values have to be aggregated. The following
1217characters are currently supported :
1218
1219 C The value is valid for a whole cluster of nodes, which is the set of nodes
1220 communicating over the peers protocol. An example could be the amount of
1221 entries present in a stick table that is replicated with other peers. At
1222 the moment no metric use this scope.
1223
1224 P The value is valid only for the process reporting it. Most metrics use
1225 this scope.
1226
1227 S The value is valid for the whole service, which is the set of processes
1228 started together from the same configuration file. All metrics originating
1229 from the configuration use this scope. Some other metrics may use it as
1230 well for some shared resources (eg: shared SSL cache statistics).
1231
1232 s The value is valid for the whole system, such as the system's hostname,
1233 current date or resource usage. At the moment this scope is not used by
1234 any metric.
1235
1236Consumers of these information will generally have enough of these 3 characters
1237to determine how to accurately report aggregated information across multiple
1238processes.
1239
1240After this column, the third column indicates the type of the field, among "s32"
1241(signed 32-bit integer), "s64" (signed 64-bit integer), "u32" (unsigned 32-bit
1242integer), "u64" (unsigned 64-bit integer), "str" (string). It is important to
1243know the type before parsing the value in order to properly read it. For example
1244a string containing only digits is still a string an not an integer (eg: an
1245error code extracted by a check).
1246
1247Then the fourth column is the value itself, encoded according to its type.
1248Strings are dumped as-is immediately after the colon without any leading space.
1249If a string contains a colon, it will appear normally. This means that the
1250output should not be exclusively split around colons or some check outputs
1251or server addresses might be truncated.
1252
1253
12549.3. Unix Socket commands
Willy Tarreau44aed902015-10-13 14:45:29 +02001255-------------------------
1256
1257The stats socket is not enabled by default. In order to enable it, it is
1258necessary to add one line in the global section of the haproxy configuration.
1259A second line is recommended to set a larger timeout, always appreciated when
1260issuing commands by hand :
1261
1262 global
1263 stats socket /var/run/haproxy.sock mode 600 level admin
1264 stats timeout 2m
1265
1266It is also possible to add multiple instances of the stats socket by repeating
1267the line, and make them listen to a TCP port instead of a UNIX socket. This is
1268never done by default because this is dangerous, but can be handy in some
1269situations :
1270
1271 global
1272 stats socket /var/run/haproxy.sock mode 600 level admin
1273 stats socket ipv4@192.168.0.1:9999 level admin
1274 stats timeout 2m
1275
1276To access the socket, an external utility such as "socat" is required. Socat is
1277a swiss-army knife to connect anything to anything. We use it to connect
1278terminals to the socket, or a couple of stdin/stdout pipes to it for scripts.
1279The two main syntaxes we'll use are the following :
1280
1281 # socat /var/run/haproxy.sock stdio
1282 # socat /var/run/haproxy.sock readline
1283
1284The first one is used with scripts. It is possible to send the output of a
1285script to haproxy, and pass haproxy's output to another script. That's useful
1286for retrieving counters or attack traces for example.
1287
1288The second one is only useful for issuing commands by hand. It has the benefit
1289that the terminal is handled by the readline library which supports line
1290editing and history, which is very convenient when issuing repeated commands
1291(eg: watch a counter).
1292
1293The socket supports two operation modes :
1294 - interactive
1295 - non-interactive
1296
1297The non-interactive mode is the default when socat connects to the socket. In
1298this mode, a single line may be sent. It is processed as a whole, responses are
1299sent back, and the connection closes after the end of the response. This is the
1300mode that scripts and monitoring tools use. It is possible to send multiple
1301commands in this mode, they need to be delimited by a semi-colon (';'). For
1302example :
1303
1304 # echo "show info;show stat;show table" | socat /var/run/haproxy stdio
1305
Dragan Dosena1c35ab2016-11-24 11:33:12 +01001306If a command needs to use a semi-colon or a backslash (eg: in a value), it
Joseph Herlant71b4b152018-11-13 16:55:16 -08001307must be preceded by a backslash ('\').
Chad Lavoiee3f50312016-05-26 16:42:25 -04001308
Willy Tarreau44aed902015-10-13 14:45:29 +02001309The interactive mode displays a prompt ('>') and waits for commands to be
1310entered on the line, then processes them, and displays the prompt again to wait
1311for a new command. This mode is entered via the "prompt" command which must be
1312sent on the first line in non-interactive mode. The mode is a flip switch, if
1313"prompt" is sent in interactive mode, it is disabled and the connection closes
1314after processing the last command of the same line.
1315
1316For this reason, when debugging by hand, it's quite common to start with the
1317"prompt" command :
1318
1319 # socat /var/run/haproxy readline
1320 prompt
1321 > show info
1322 ...
1323 >
1324
1325Since multiple commands may be issued at once, haproxy uses the empty line as a
1326delimiter to mark an end of output for each command, and takes care of ensuring
1327that no command can emit an empty line on output. A script can thus easily
1328parse the output even when multiple commands were pipelined on a single line.
1329
Aurélien Nephtaliabbf6072018-04-18 13:26:46 +02001330Some commands may take an optional payload. To add one to a command, the first
1331line needs to end with the "<<\n" pattern. The next lines will be treated as
1332the payload and can contain as many lines as needed. To validate a command with
1333a payload, it needs to end with an empty line.
1334
1335Limitations do exist: the length of the whole buffer passed to the CLI must
1336not be greater than tune.bfsize and the pattern "<<" must not be glued to the
1337last word of the line.
1338
1339When entering a paylod while in interactive mode, the prompt will change from
1340"> " to "+ ".
1341
Willy Tarreau44aed902015-10-13 14:45:29 +02001342It is important to understand that when multiple haproxy processes are started
1343on the same sockets, any process may pick up the request and will output its
1344own stats.
1345
1346The list of commands currently supported on the stats socket is provided below.
1347If an unknown command is sent, haproxy displays the usage message which reminds
1348all supported commands. Some commands support a more complex syntax, generally
1349it will explain what part of the command is invalid when this happens.
1350
Olivier Doucetd8703e82017-08-31 11:05:10 +02001351Some commands require a higher level of privilege to work. If you do not have
1352enough privilege, you will get an error "Permission denied". Please check
1353the "level" option of the "bind" keyword lines in the configuration manual
1354for more information.
1355
Willy Tarreau44aed902015-10-13 14:45:29 +02001356add acl <acl> <pattern>
1357 Add an entry into the acl <acl>. <acl> is the #<id> or the <file> returned by
1358 "show acl". This command does not verify if the entry already exists. This
1359 command cannot be used if the reference <acl> is a file also used with a map.
1360 In this case, you must use the command "add map" in place of "add acl".
1361
1362add map <map> <key> <value>
Aurélien Nephtali25650ce2018-04-18 14:04:47 +02001363add map <map> <payload>
Willy Tarreau44aed902015-10-13 14:45:29 +02001364 Add an entry into the map <map> to associate the value <value> to the key
1365 <key>. This command does not verify if the entry already exists. It is
1366 mainly used to fill a map after a clear operation. Note that if the reference
1367 <map> is a file and is shared with a map, this map will contain also a new
Aurélien Nephtali25650ce2018-04-18 14:04:47 +02001368 pattern entry. Using the payload syntax it is possible to add multiple
1369 key/value pairs by entering them on separate lines. On each new line, the
1370 first word is the key and the rest of the line is considered to be the value
1371 which can even contains spaces.
1372
1373 Example:
1374
1375 # socat /tmp/sock1 -
1376 prompt
1377
1378 > add map #-1 <<
1379 + key1 value1
1380 + key2 value2 with spaces
1381 + key3 value3 also with spaces
1382 + key4 value4
1383
1384 >
Willy Tarreau44aed902015-10-13 14:45:29 +02001385
1386clear counters
1387 Clear the max values of the statistics counters in each proxy (frontend &
Willy Tarreaud80cb4e2018-01-20 19:30:13 +01001388 backend) and in each server. The accumulated counters are not affected. The
1389 internal activity counters reported by "show activity" are also reset. This
Willy Tarreau44aed902015-10-13 14:45:29 +02001390 can be used to get clean counters after an incident, without having to
1391 restart nor to clear traffic counters. This command is restricted and can
1392 only be issued on sockets configured for levels "operator" or "admin".
1393
1394clear counters all
1395 Clear all statistics counters in each proxy (frontend & backend) and in each
1396 server. This has the same effect as restarting. This command is restricted
1397 and can only be issued on sockets configured for level "admin".
1398
1399clear acl <acl>
1400 Remove all entries from the acl <acl>. <acl> is the #<id> or the <file>
1401 returned by "show acl". Note that if the reference <acl> is a file and is
1402 shared with a map, this map will be also cleared.
1403
1404clear map <map>
1405 Remove all entries from the map <map>. <map> is the #<id> or the <file>
1406 returned by "show map". Note that if the reference <map> is a file and is
1407 shared with a acl, this acl will be also cleared.
1408
1409clear table <table> [ data.<type> <operator> <value> ] | [ key <key> ]
1410 Remove entries from the stick-table <table>.
1411
1412 This is typically used to unblock some users complaining they have been
1413 abusively denied access to a service, but this can also be used to clear some
1414 stickiness entries matching a server that is going to be replaced (see "show
1415 table" below for details). Note that sometimes, removal of an entry will be
1416 refused because it is currently tracked by a session. Retrying a few seconds
1417 later after the session ends is usual enough.
1418
1419 In the case where no options arguments are given all entries will be removed.
1420
1421 When the "data." form is used entries matching a filter applied using the
1422 stored data (see "stick-table" in section 4.2) are removed. A stored data
1423 type must be specified in <type>, and this data type must be stored in the
1424 table otherwise an error is reported. The data is compared according to
1425 <operator> with the 64-bit integer <value>. Operators are the same as with
1426 the ACLs :
1427
1428 - eq : match entries whose data is equal to this value
1429 - ne : match entries whose data is not equal to this value
1430 - le : match entries whose data is less than or equal to this value
1431 - ge : match entries whose data is greater than or equal to this value
1432 - lt : match entries whose data is less than this value
1433 - gt : match entries whose data is greater than this value
1434
1435 When the key form is used the entry <key> is removed. The key must be of the
1436 same type as the table, which currently is limited to IPv4, IPv6, integer and
1437 string.
1438
1439 Example :
1440 $ echo "show table http_proxy" | socat stdio /tmp/sock1
1441 >>> # table: http_proxy, type: ip, size:204800, used:2
1442 >>> 0x80e6a4c: key=127.0.0.1 use=0 exp=3594729 gpc0=0 conn_rate(30000)=1 \
1443 bytes_out_rate(60000)=187
1444 >>> 0x80e6a80: key=127.0.0.2 use=0 exp=3594740 gpc0=1 conn_rate(30000)=10 \
1445 bytes_out_rate(60000)=191
1446
1447 $ echo "clear table http_proxy key 127.0.0.1" | socat stdio /tmp/sock1
1448
1449 $ echo "show table http_proxy" | socat stdio /tmp/sock1
1450 >>> # table: http_proxy, type: ip, size:204800, used:1
1451 >>> 0x80e6a80: key=127.0.0.2 use=0 exp=3594740 gpc0=1 conn_rate(30000)=10 \
1452 bytes_out_rate(60000)=191
1453 $ echo "clear table http_proxy data.gpc0 eq 1" | socat stdio /tmp/sock1
1454 $ echo "show table http_proxy" | socat stdio /tmp/sock1
1455 >>> # table: http_proxy, type: ip, size:204800, used:1
1456
1457del acl <acl> [<key>|#<ref>]
1458 Delete all the acl entries from the acl <acl> corresponding to the key <key>.
1459 <acl> is the #<id> or the <file> returned by "show acl". If the <ref> is used,
1460 this command delete only the listed reference. The reference can be found with
1461 listing the content of the acl. Note that if the reference <acl> is a file and
1462 is shared with a map, the entry will be also deleted in the map.
1463
1464del map <map> [<key>|#<ref>]
1465 Delete all the map entries from the map <map> corresponding to the key <key>.
1466 <map> is the #<id> or the <file> returned by "show map". If the <ref> is used,
1467 this command delete only the listed reference. The reference can be found with
1468 listing the content of the map. Note that if the reference <map> is a file and
1469 is shared with a acl, the entry will be also deleted in the map.
1470
1471disable agent <backend>/<server>
1472 Mark the auxiliary agent check as temporarily stopped.
1473
1474 In the case where an agent check is being run as a auxiliary check, due
1475 to the agent-check parameter of a server directive, new checks are only
Dan Lloyd8e48b872016-07-01 21:01:18 -04001476 initialized when the agent is in the enabled. Thus, disable agent will
Willy Tarreau44aed902015-10-13 14:45:29 +02001477 prevent any new agent checks from begin initiated until the agent
1478 re-enabled using enable agent.
1479
1480 When an agent is disabled the processing of an auxiliary agent check that
1481 was initiated while the agent was set as enabled is as follows: All
1482 results that would alter the weight, specifically "drain" or a weight
1483 returned by the agent, are ignored. The processing of agent check is
1484 otherwise unchanged.
1485
1486 The motivation for this feature is to allow the weight changing effects
1487 of the agent checks to be paused to allow the weight of a server to be
1488 configured using set weight without being overridden by the agent.
1489
1490 This command is restricted and can only be issued on sockets configured for
1491 level "admin".
1492
Olivier Houchard614f8d72017-03-14 20:08:46 +01001493disable dynamic-cookie backend <backend>
1494 Disable the generation of dynamic cookies fot the backend <backend>
1495
Willy Tarreau44aed902015-10-13 14:45:29 +02001496disable frontend <frontend>
1497 Mark the frontend as temporarily stopped. This corresponds to the mode which
1498 is used during a soft restart : the frontend releases the port but can be
1499 enabled again if needed. This should be used with care as some non-Linux OSes
1500 are unable to enable it back. This is intended to be used in environments
1501 where stopping a proxy is not even imaginable but a misconfigured proxy must
1502 be fixed. That way it's possible to release the port and bind it into another
1503 process to restore operations. The frontend will appear with status "STOP"
1504 on the stats page.
1505
1506 The frontend may be specified either by its name or by its numeric ID,
1507 prefixed with a sharp ('#').
1508
1509 This command is restricted and can only be issued on sockets configured for
1510 level "admin".
1511
1512disable health <backend>/<server>
1513 Mark the primary health check as temporarily stopped. This will disable
1514 sending of health checks, and the last health check result will be ignored.
1515 The server will be in unchecked state and considered UP unless an auxiliary
1516 agent check forces it down.
1517
1518 This command is restricted and can only be issued on sockets configured for
1519 level "admin".
1520
1521disable server <backend>/<server>
1522 Mark the server DOWN for maintenance. In this mode, no more checks will be
1523 performed on the server until it leaves maintenance.
1524 If the server is tracked by other servers, those servers will be set to DOWN
1525 during the maintenance.
1526
1527 In the statistics page, a server DOWN for maintenance will appear with a
1528 "MAINT" status, its tracking servers with the "MAINT(via)" one.
1529
1530 Both the backend and the server may be specified either by their name or by
1531 their numeric ID, prefixed with a sharp ('#').
1532
1533 This command is restricted and can only be issued on sockets configured for
1534 level "admin".
1535
1536enable agent <backend>/<server>
1537 Resume auxiliary agent check that was temporarily stopped.
1538
1539 See "disable agent" for details of the effect of temporarily starting
1540 and stopping an auxiliary agent.
1541
1542 This command is restricted and can only be issued on sockets configured for
1543 level "admin".
1544
Olivier Houchard614f8d72017-03-14 20:08:46 +01001545enable dynamic-cookie backend <backend>
1546 Enable the generation of dynamic cookies fot the backend <backend>
1547 A secret key must also be provided
1548
Willy Tarreau44aed902015-10-13 14:45:29 +02001549enable frontend <frontend>
1550 Resume a frontend which was temporarily stopped. It is possible that some of
1551 the listening ports won't be able to bind anymore (eg: if another process
1552 took them since the 'disable frontend' operation). If this happens, an error
1553 is displayed. Some operating systems might not be able to resume a frontend
1554 which was disabled.
1555
1556 The frontend may be specified either by its name or by its numeric ID,
1557 prefixed with a sharp ('#').
1558
1559 This command is restricted and can only be issued on sockets configured for
1560 level "admin".
1561
1562enable health <backend>/<server>
1563 Resume a primary health check that was temporarily stopped. This will enable
1564 sending of health checks again. Please see "disable health" for details.
1565
1566 This command is restricted and can only be issued on sockets configured for
1567 level "admin".
1568
1569enable server <backend>/<server>
1570 If the server was previously marked as DOWN for maintenance, this marks the
1571 server UP and checks are re-enabled.
1572
1573 Both the backend and the server may be specified either by their name or by
1574 their numeric ID, prefixed with a sharp ('#').
1575
1576 This command is restricted and can only be issued on sockets configured for
1577 level "admin".
1578
1579get map <map> <value>
1580get acl <acl> <value>
1581 Lookup the value <value> in the map <map> or in the ACL <acl>. <map> or <acl>
1582 are the #<id> or the <file> returned by "show map" or "show acl". This command
1583 returns all the matching patterns associated with this map. This is useful for
1584 debugging maps and ACLs. The output format is composed by one line par
1585 matching type. Each line is composed by space-delimited series of words.
1586
1587 The first two words are:
1588
1589 <match method>: The match method applied. It can be "found", "bool",
1590 "int", "ip", "bin", "len", "str", "beg", "sub", "dir",
1591 "dom", "end" or "reg".
1592
1593 <match result>: The result. Can be "match" or "no-match".
1594
1595 The following words are returned only if the pattern matches an entry.
1596
1597 <index type>: "tree" or "list". The internal lookup algorithm.
1598
1599 <case>: "case-insensitive" or "case-sensitive". The
1600 interpretation of the case.
1601
1602 <entry matched>: match="<entry>". Return the matched pattern. It is
1603 useful with regular expressions.
1604
1605 The two last word are used to show the returned value and its type. With the
1606 "acl" case, the pattern doesn't exist.
1607
1608 return=nothing: No return because there are no "map".
1609 return="<value>": The value returned in the string format.
1610 return=cannot-display: The value cannot be converted as string.
1611
1612 type="<type>": The type of the returned sample.
1613
1614get weight <backend>/<server>
1615 Report the current weight and the initial weight of server <server> in
1616 backend <backend> or an error if either doesn't exist. The initial weight is
1617 the one that appears in the configuration file. Both are normally equal
1618 unless the current weight has been changed. Both the backend and the server
1619 may be specified either by their name or by their numeric ID, prefixed with a
1620 sharp ('#').
1621
1622help
1623 Print the list of known keywords and their basic usage. The same help screen
1624 is also displayed for unknown commands.
1625
1626prompt
1627 Toggle the prompt at the beginning of the line and enter or leave interactive
1628 mode. In interactive mode, the connection is not closed after a command
1629 completes. Instead, the prompt will appear again, indicating the user that
1630 the interpreter is waiting for a new command. The prompt consists in a right
1631 angle bracket followed by a space "> ". This mode is particularly convenient
1632 when one wants to periodically check information such as stats or errors.
1633 It is also a good idea to enter interactive mode before issuing a "help"
1634 command.
1635
1636quit
1637 Close the connection when in interactive mode.
1638
Olivier Houchard614f8d72017-03-14 20:08:46 +01001639set dynamic-cookie-key backend <backend> <value>
1640 Modify the secret key used to generate the dynamic persistent cookies.
1641 This will break the existing sessions.
1642
Willy Tarreau44aed902015-10-13 14:45:29 +02001643set map <map> [<key>|#<ref>] <value>
1644 Modify the value corresponding to each key <key> in a map <map>. <map> is the
1645 #<id> or <file> returned by "show map". If the <ref> is used in place of
1646 <key>, only the entry pointed by <ref> is changed. The new value is <value>.
1647
1648set maxconn frontend <frontend> <value>
1649 Dynamically change the specified frontend's maxconn setting. Any positive
1650 value is allowed including zero, but setting values larger than the global
1651 maxconn does not make much sense. If the limit is increased and connections
1652 were pending, they will immediately be accepted. If it is lowered to a value
1653 below the current number of connections, new connections acceptation will be
1654 delayed until the threshold is reached. The frontend might be specified by
1655 either its name or its numeric ID prefixed with a sharp ('#').
1656
Andrew Hayworthedb93a72015-10-27 21:46:25 +00001657set maxconn server <backend/server> <value>
1658 Dynamically change the specified server's maxconn setting. Any positive
1659 value is allowed including zero, but setting values larger than the global
1660 maxconn does not make much sense.
1661
Willy Tarreau44aed902015-10-13 14:45:29 +02001662set maxconn global <maxconn>
1663 Dynamically change the global maxconn setting within the range defined by the
1664 initial global maxconn setting. If it is increased and connections were
1665 pending, they will immediately be accepted. If it is lowered to a value below
1666 the current number of connections, new connections acceptation will be
1667 delayed until the threshold is reached. A value of zero restores the initial
1668 setting.
1669
Willy Tarreau75c62c22018-11-22 11:02:09 +01001670set profiling { tasks } { on | off }
1671 Enables or disables CPU profiling for the indicated subsystem. This is
1672 equivalent to setting or clearing the "profiling" settings in the "global"
1673 section of the configuration file. Please also see "show profiling".
1674
Willy Tarreau44aed902015-10-13 14:45:29 +02001675set rate-limit connections global <value>
1676 Change the process-wide connection rate limit, which is set by the global
1677 'maxconnrate' setting. A value of zero disables the limitation. This limit
1678 applies to all frontends and the change has an immediate effect. The value
1679 is passed in number of connections per second.
1680
1681set rate-limit http-compression global <value>
1682 Change the maximum input compression rate, which is set by the global
1683 'maxcomprate' setting. A value of zero disables the limitation. The value is
1684 passed in number of kilobytes per second. The value is available in the "show
1685 info" on the line "CompressBpsRateLim" in bytes.
1686
1687set rate-limit sessions global <value>
1688 Change the process-wide session rate limit, which is set by the global
1689 'maxsessrate' setting. A value of zero disables the limitation. This limit
1690 applies to all frontends and the change has an immediate effect. The value
1691 is passed in number of sessions per second.
1692
1693set rate-limit ssl-sessions global <value>
1694 Change the process-wide SSL session rate limit, which is set by the global
1695 'maxsslrate' setting. A value of zero disables the limitation. This limit
1696 applies to all frontends and the change has an immediate effect. The value
1697 is passed in number of sessions per second sent to the SSL stack. It applies
1698 before the handshake in order to protect the stack against handshake abuses.
1699
Baptiste Assmann3749ebf2016-08-03 22:34:12 +02001700set server <backend>/<server> addr <ip4 or ip6 address> [port <port>]
Willy Tarreau44aed902015-10-13 14:45:29 +02001701 Replace the current IP address of a server by the one provided.
Baptiste Assmann3749ebf2016-08-03 22:34:12 +02001702 Optionnaly, the port can be changed using the 'port' parameter.
1703 Note that changing the port also support switching from/to port mapping
1704 (notation with +X or -Y), only if a port is configured for the health check.
Willy Tarreau44aed902015-10-13 14:45:29 +02001705
1706set server <backend>/<server> agent [ up | down ]
1707 Force a server's agent to a new state. This can be useful to immediately
1708 switch a server's state regardless of some slow agent checks for example.
1709 Note that the change is propagated to tracking servers if any.
1710
Misiek43972902017-01-09 09:53:06 +01001711set server <backend>/<server> agent-addr <addr>
1712 Change addr for servers agent checks. Allows to migrate agent-checks to
1713 another address at runtime. You can specify both IP and hostname, it will be
1714 resolved.
1715
1716set server <backend>/<server> agent-send <value>
1717 Change agent string sent to agent check target. Allows to update string while
1718 changing server address to keep those two matching.
1719
Willy Tarreau44aed902015-10-13 14:45:29 +02001720set server <backend>/<server> health [ up | stopping | down ]
1721 Force a server's health to a new state. This can be useful to immediately
1722 switch a server's state regardless of some slow health checks for example.
1723 Note that the change is propagated to tracking servers if any.
1724
Baptiste Assmann50946562016-08-31 23:26:29 +02001725set server <backend>/<server> check-port <port>
1726 Change the port used for health checking to <port>
1727
Willy Tarreau44aed902015-10-13 14:45:29 +02001728set server <backend>/<server> state [ ready | drain | maint ]
1729 Force a server's administrative state to a new state. This can be useful to
1730 disable load balancing and/or any traffic to a server. Setting the state to
1731 "ready" puts the server in normal mode, and the command is the equivalent of
1732 the "enable server" command. Setting the state to "maint" disables any traffic
1733 to the server as well as any health checks. This is the equivalent of the
1734 "disable server" command. Setting the mode to "drain" only removes the server
1735 from load balancing but still allows it to be checked and to accept new
1736 persistent connections. Changes are propagated to tracking servers if any.
1737
1738set server <backend>/<server> weight <weight>[%]
1739 Change a server's weight to the value passed in argument. This is the exact
1740 equivalent of the "set weight" command below.
1741
Frédéric Lécailleb418c122017-04-26 11:24:02 +02001742set server <backend>/<server> fqdn <FQDN>
Lukas Tribusc5dd5a52018-08-14 11:39:35 +02001743 Change a server's FQDN to the value passed in argument. This requires the
1744 internal run-time DNS resolver to be configured and enabled for this server.
Frédéric Lécailleb418c122017-04-26 11:24:02 +02001745
Andjelko Iharosc4df59e2017-07-20 11:59:48 +02001746set severity-output [ none | number | string ]
1747 Change the severity output format of the stats socket connected to for the
1748 duration of the current session.
1749
Aurélien Nephtali1e0867c2018-04-18 14:04:58 +02001750set ssl ocsp-response <response | payload>
Willy Tarreau44aed902015-10-13 14:45:29 +02001751 This command is used to update an OCSP Response for a certificate (see "crt"
1752 on "bind" lines). Same controls are performed as during the initial loading of
1753 the response. The <response> must be passed as a base64 encoded string of the
Emmanuel Hocdet2c32d8f2017-05-22 14:58:00 +02001754 DER encoded response from the OCSP server. This command is not supported with
1755 BoringSSL.
Willy Tarreau44aed902015-10-13 14:45:29 +02001756
1757 Example:
1758 openssl ocsp -issuer issuer.pem -cert server.pem \
1759 -host ocsp.issuer.com:80 -respout resp.der
1760 echo "set ssl ocsp-response $(base64 -w 10000 resp.der)" | \
1761 socat stdio /var/run/haproxy.stat
1762
Aurélien Nephtali1e0867c2018-04-18 14:04:58 +02001763 using the payload syntax:
1764 echo -e "set ssl ocsp-response <<\n$(base64 resp.der)\n" | \
1765 socat stdio /var/run/haproxy.stat
1766
Willy Tarreau44aed902015-10-13 14:45:29 +02001767set ssl tls-key <id> <tlskey>
1768 Set the next TLS key for the <id> listener to <tlskey>. This key becomes the
1769 ultimate key, while the penultimate one is used for encryption (others just
1770 decrypt). The oldest TLS key present is overwritten. <id> is either a numeric
1771 #<id> or <file> returned by "show tls-keys". <tlskey> is a base64 encoded 48
Emeric Brun9e754772019-01-10 17:51:55 +01001772 or 80 bits TLS ticket key (ex. openssl rand 80 | openssl base64 -A).
Willy Tarreau44aed902015-10-13 14:45:29 +02001773
1774set table <table> key <key> [data.<data_type> <value>]*
1775 Create or update a stick-table entry in the table. If the key is not present,
1776 an entry is inserted. See stick-table in section 4.2 to find all possible
1777 values for <data_type>. The most likely use consists in dynamically entering
1778 entries for source IP addresses, with a flag in gpc0 to dynamically block an
1779 IP address or affect its quality of service. It is possible to pass multiple
1780 data_types in a single call.
1781
1782set timeout cli <delay>
1783 Change the CLI interface timeout for current connection. This can be useful
1784 during long debugging sessions where the user needs to constantly inspect
1785 some indicators without being disconnected. The delay is passed in seconds.
1786
1787set weight <backend>/<server> <weight>[%]
1788 Change a server's weight to the value passed in argument. If the value ends
1789 with the '%' sign, then the new weight will be relative to the initially
1790 configured weight. Absolute weights are permitted between 0 and 256.
1791 Relative weights must be positive with the resulting absolute weight is
1792 capped at 256. Servers which are part of a farm running a static
1793 load-balancing algorithm have stricter limitations because the weight
1794 cannot change once set. Thus for these servers, the only accepted values
1795 are 0 and 100% (or 0 and the initial weight). Changes take effect
1796 immediately, though certain LB algorithms require a certain amount of
1797 requests to consider changes. A typical usage of this command is to
1798 disable a server during an update by setting its weight to zero, then to
1799 enable it again after the update by setting it back to 100%. This command
1800 is restricted and can only be issued on sockets configured for level
1801 "admin". Both the backend and the server may be specified either by their
1802 name or by their numeric ID, prefixed with a sharp ('#').
1803
Willy Tarreaud6129fc2017-07-28 16:52:23 +02001804show acl [<acl>]
1805 Dump info about acl converters. Without argument, the list of all available
1806 acls is returned. If a <acl> is specified, its contents are dumped. <acl> if
1807 the #<id> or <file>. The dump format is the same than the map even for the
1808 sample value. The data returned are not a list of available ACL, but are the
1809 list of all patterns composing any ACL. Many of these patterns can be shared
1810 with maps.
1811
1812show backend
1813 Dump the list of backends available in the running process
1814
William Lallemand67a234f2018-12-13 09:05:45 +01001815show cli level
1816 Display the CLI level of the current CLI session. The result could be
1817 'admin', 'operator' or 'user'. See also the 'operator' and 'user' commands.
1818
1819 Example :
1820
1821 $ socat /tmp/sock1 readline
1822 prompt
1823 > operator
1824 > show cli level
1825 operator
1826 > user
1827 > show cli level
1828 user
1829 > operator
1830 Permission denied
1831
1832operator
1833 Decrease the CLI level of the current CLI session to operator. It can't be
1834 increase. See also "show cli level"
1835
1836user
1837 Decrease the CLI level of the current CLI session to user. It can't be
1838 increase. See also "show cli level"
1839
William Lallemand51132162016-12-16 16:38:58 +01001840show cli sockets
1841 List CLI sockets. The output format is composed of 3 fields separated by
1842 spaces. The first field is the socket address, it can be a unix socket, a
1843 ipv4 address:port couple or a ipv6 one. Socket of other types won't be dump.
1844 The second field describe the level of the socket: 'admin', 'user' or
1845 'operator'. The last field list the processes on which the socket is bound,
1846 separated by commas, it can be numbers or 'all'.
1847
1848 Example :
1849
1850 $ echo 'show cli sockets' | socat stdio /tmp/sock1
1851 # socket lvl processes
1852 /tmp/sock1 admin all
1853 127.0.0.1:9999 user 2,3,4
1854 127.0.0.2:9969 user 2
1855 [::1]:9999 operator 2
1856
William Lallemand86d0df02017-11-24 21:36:45 +01001857show cache
Cyril Bonté7b888f12017-11-26 22:24:31 +01001858 List the configured caches and the objects stored in each cache tree.
William Lallemand86d0df02017-11-24 21:36:45 +01001859
1860 $ echo 'show cache' | socat stdio /tmp/sock1
1861 0x7f6ac6c5b03a: foobar (shctx:0x7f6ac6c5b000, available blocks:3918)
1862 1 2 3 4
1863
1864 1. pointer to the cache structure
1865 2. cache name
1866 3. pointer to the mmap area (shctx)
1867 4. number of blocks available for reuse in the shctx
1868
1869 0x7f6ac6c5b4cc hash:286881868 size:39114 (39 blocks), refcount:9, expire:237
1870 1 2 3 4 5 6
1871
1872 1. pointer to the cache entry
1873 2. first 32 bits of the hash
1874 3. size of the object in bytes
1875 4. number of blocks used for the object
1876 5. number of transactions using the entry
1877 6. expiration time, can be negative if already expired
1878
Willy Tarreauae795722016-02-16 11:27:28 +01001879show env [<name>]
1880 Dump one or all environment variables known by the process. Without any
1881 argument, all variables are dumped. With an argument, only the specified
1882 variable is dumped if it exists. Otherwise "Variable not found" is emitted.
1883 Variables are dumped in the same format as they are stored or returned by the
1884 "env" utility, that is, "<name>=<value>". This can be handy when debugging
1885 certain configuration files making heavy use of environment variables to
1886 ensure that they contain the expected values. This command is restricted and
1887 can only be issued on sockets configured for levels "operator" or "admin".
1888
Willy Tarreau35069f82016-11-25 09:16:37 +01001889show errors [<iid>|<proxy>] [request|response]
Willy Tarreau44aed902015-10-13 14:45:29 +02001890 Dump last known request and response errors collected by frontends and
1891 backends. If <iid> is specified, the limit the dump to errors concerning
Willy Tarreau234ba2d2016-11-25 08:39:10 +01001892 either frontend or backend whose ID is <iid>. Proxy ID "-1" will cause
1893 all instances to be dumped. If a proxy name is specified instead, its ID
Willy Tarreau35069f82016-11-25 09:16:37 +01001894 will be used as the filter. If "request" or "response" is added after the
1895 proxy name or ID, only request or response errors will be dumped. This
1896 command is restricted and can only be issued on sockets configured for
1897 levels "operator" or "admin".
Willy Tarreau44aed902015-10-13 14:45:29 +02001898
1899 The errors which may be collected are the last request and response errors
1900 caused by protocol violations, often due to invalid characters in header
1901 names. The report precisely indicates what exact character violated the
1902 protocol. Other important information such as the exact date the error was
1903 detected, frontend and backend names, the server name (when known), the
1904 internal session ID and the source address which has initiated the session
1905 are reported too.
1906
1907 All characters are returned, and non-printable characters are encoded. The
1908 most common ones (\t = 9, \n = 10, \r = 13 and \e = 27) are encoded as one
1909 letter following a backslash. The backslash itself is encoded as '\\' to
1910 avoid confusion. Other non-printable characters are encoded '\xNN' where
1911 NN is the two-digits hexadecimal representation of the character's ASCII
1912 code.
1913
1914 Lines are prefixed with the position of their first character, starting at 0
1915 for the beginning of the buffer. At most one input line is printed per line,
1916 and large lines will be broken into multiple consecutive output lines so that
1917 the output never goes beyond 79 characters wide. It is easy to detect if a
1918 line was broken, because it will not end with '\n' and the next line's offset
1919 will be followed by a '+' sign, indicating it is a continuation of previous
1920 line.
1921
1922 Example :
Willy Tarreau35069f82016-11-25 09:16:37 +01001923 $ echo "show errors -1 response" | socat stdio /tmp/sock1
Willy Tarreau44aed902015-10-13 14:45:29 +02001924 >>> [04/Mar/2009:15:46:56.081] backend http-in (#2) : invalid response
1925 src 127.0.0.1, session #54, frontend fe-eth0 (#1), server s2 (#1)
1926 response length 213 bytes, error at position 23:
1927
1928 00000 HTTP/1.0 200 OK\r\n
1929 00017 header/bizarre:blah\r\n
1930 00038 Location: blah\r\n
1931 00054 Long-line: this is a very long line which should b
1932 00104+ e broken into multiple lines on the output buffer,
1933 00154+ otherwise it would be too large to print in a ter
1934 00204+ minal\r\n
1935 00211 \r\n
1936
1937 In the example above, we see that the backend "http-in" which has internal
1938 ID 2 has blocked an invalid response from its server s2 which has internal
1939 ID 1. The request was on session 54 initiated by source 127.0.0.1 and
1940 received by frontend fe-eth0 whose ID is 1. The total response length was
1941 213 bytes when the error was detected, and the error was at byte 23. This
1942 is the slash ('/') in header name "header/bizarre", which is not a valid
1943 HTTP character for a header name.
1944
Willy Tarreau7a4a0ac2017-07-25 19:32:50 +02001945show fd [<fd>]
1946 Dump the list of either all open file descriptors or just the one number <fd>
1947 if specified. This is only aimed at developers who need to observe internal
1948 states in order to debug complex issues such as abnormal CPU usages. One fd
1949 is reported per lines, and for each of them, its state in the poller using
1950 upper case letters for enabled flags and lower case for disabled flags, using
1951 "P" for "polled", "R" for "ready", "A" for "active", the events status using
1952 "H" for "hangup", "E" for "error", "O" for "output", "P" for "priority" and
1953 "I" for "input", a few other flags like "N" for "new" (just added into the fd
1954 cache), "U" for "updated" (received an update in the fd cache), "L" for
1955 "linger_risk", "C" for "cloned", then the cached entry position, the pointer
1956 to the internal owner, the pointer to the I/O callback and its name when
1957 known. When the owner is a connection, the connection flags, and the target
1958 are reported (frontend, proxy or server). When the owner is a listener, the
1959 listener's state and its frontend are reported. There is no point in using
1960 this command without a good knowledge of the internals. It's worth noting
1961 that the output format may evolve over time so this output must not be parsed
1962 by tools designed to be durable.
1963
Willy Tarreaud80cb4e2018-01-20 19:30:13 +01001964show activity
1965 Reports some counters about internal events that will help developers and
1966 more generally people who know haproxy well enough to narrow down the causes
1967 of reports of abnormal behaviours. A typical example would be a properly
1968 running process never sleeping and eating 100% of the CPU. The output fields
1969 will be made of one line per metric, and per-thread counters on the same
1970 line. These counters are 32-bit and will wrap during the process' life, which
1971 is not a problem since calls to this command will typically be performed
1972 twice. The fields are purposely not documented so that their exact meaning is
1973 verified in the code where the counters are fed. These values are also reset
1974 by the "clear counters" command.
1975
Simon Horman05ee2132017-01-04 09:37:25 +01001976show info [typed|json]
Willy Tarreau5d8b9792016-03-11 11:09:34 +01001977 Dump info about haproxy status on current process. If "typed" is passed as an
1978 optional argument, field numbers, names and types are emitted as well so that
1979 external monitoring products can easily retrieve, possibly aggregate, then
1980 report information found in fields they don't know. Each field is dumped on
Simon Horman05ee2132017-01-04 09:37:25 +01001981 its own line. If "json" is passed as an optional argument then
1982 information provided by "typed" output is provided in JSON format as a
1983 list of JSON objects. By default, the format contains only two columns
1984 delimited by a colon (':'). The left one is the field name and the right
1985 one is the value. It is very important to note that in typed output
1986 format, the dump for a single object is contiguous so that there is no
1987 need for a consumer to store everything at once.
Willy Tarreau5d8b9792016-03-11 11:09:34 +01001988
1989 When using the typed output format, each line is made of 4 columns delimited
1990 by colons (':'). The first column is a dot-delimited series of 3 elements. The
1991 first element is the numeric position of the field in the list (starting at
1992 zero). This position shall not change over time, but holes are to be expected,
1993 depending on build options or if some fields are deleted in the future. The
1994 second element is the field name as it appears in the default "show info"
1995 output. The third element is the relative process number starting at 1.
1996
1997 The rest of the line starting after the first colon follows the "typed output
1998 format" described in the section above. In short, the second column (after the
1999 first ':') indicates the origin, nature and scope of the variable. The third
2000 column indicates the type of the field, among "s32", "s64", "u32", "u64" and
2001 "str". Then the fourth column is the value itself, which the consumer knows
2002 how to parse thanks to column 3 and how to process thanks to column 2.
2003
2004 Thus the overall line format in typed mode is :
2005
2006 <field_pos>.<field_name>.<process_num>:<tags>:<type>:<value>
2007
2008 Example :
2009
2010 > show info
2011 Name: HAProxy
2012 Version: 1.7-dev1-de52ea-146
2013 Release_date: 2016/03/11
2014 Nbproc: 1
2015 Process_num: 1
2016 Pid: 28105
2017 Uptime: 0d 0h00m04s
2018 Uptime_sec: 4
2019 Memmax_MB: 0
2020 PoolAlloc_MB: 0
2021 PoolUsed_MB: 0
2022 PoolFailed: 0
2023 (...)
2024
2025 > show info typed
2026 0.Name.1:POS:str:HAProxy
2027 1.Version.1:POS:str:1.7-dev1-de52ea-146
2028 2.Release_date.1:POS:str:2016/03/11
2029 3.Nbproc.1:CGS:u32:1
2030 4.Process_num.1:KGP:u32:1
2031 5.Pid.1:SGP:u32:28105
2032 6.Uptime.1:MDP:str:0d 0h00m08s
2033 7.Uptime_sec.1:MDP:u32:8
2034 8.Memmax_MB.1:CLP:u32:0
2035 9.PoolAlloc_MB.1:MGP:u32:0
2036 10.PoolUsed_MB.1:MGP:u32:0
2037 11.PoolFailed.1:MCP:u32:0
2038 (...)
2039
Simon Horman1084a362016-11-21 17:00:24 +01002040 In the typed format, the presence of the process ID at the end of the
2041 first column makes it very easy to visually aggregate outputs from
2042 multiple processes.
Willy Tarreau5d8b9792016-03-11 11:09:34 +01002043 Example :
2044
2045 $ ( echo show info typed | socat /var/run/haproxy.sock1 ; \
2046 echo show info typed | socat /var/run/haproxy.sock2 ) | \
2047 sort -t . -k 1,1n -k 2,2 -k 3,3n
2048 0.Name.1:POS:str:HAProxy
2049 0.Name.2:POS:str:HAProxy
2050 1.Version.1:POS:str:1.7-dev1-868ab3-148
2051 1.Version.2:POS:str:1.7-dev1-868ab3-148
2052 2.Release_date.1:POS:str:2016/03/11
2053 2.Release_date.2:POS:str:2016/03/11
2054 3.Nbproc.1:CGS:u32:2
2055 3.Nbproc.2:CGS:u32:2
2056 4.Process_num.1:KGP:u32:1
2057 4.Process_num.2:KGP:u32:2
2058 5.Pid.1:SGP:u32:30120
2059 5.Pid.2:SGP:u32:30121
2060 6.Uptime.1:MDP:str:0d 0h01m28s
2061 6.Uptime.2:MDP:str:0d 0h01m28s
2062 (...)
Willy Tarreau44aed902015-10-13 14:45:29 +02002063
Simon Horman05ee2132017-01-04 09:37:25 +01002064 The format of JSON output is described in a schema which may be output
Simon Horman6f6bb382017-01-04 09:37:26 +01002065 using "show schema json".
Simon Horman05ee2132017-01-04 09:37:25 +01002066
2067 The JSON output contains no extra whitespace in order to reduce the
2068 volume of output. For human consumption passing the output through a
2069 pretty printer may be helpful. Example :
2070
2071 $ echo "show info json" | socat /var/run/haproxy.sock stdio | \
2072 python -m json.tool
2073
Simon Horman6f6bb382017-01-04 09:37:26 +01002074 The JSON output contains no extra whitespace in order to reduce the
2075 volume of output. For human consumption passing the output through a
2076 pretty printer may be helpful. Example :
2077
2078 $ echo "show info json" | socat /var/run/haproxy.sock stdio | \
2079 python -m json.tool
2080
Willy Tarreau44aed902015-10-13 14:45:29 +02002081show map [<map>]
2082 Dump info about map converters. Without argument, the list of all available
2083 maps is returned. If a <map> is specified, its contents are dumped. <map> is
2084 the #<id> or <file>. The first column is a unique identifier. It can be used
2085 as reference for the operation "del map" and "set map". The second column is
2086 the pattern and the third column is the sample if available. The data returned
2087 are not directly a list of available maps, but are the list of all patterns
2088 composing any map. Many of these patterns can be shared with ACL.
2089
Willy Tarreau44aed902015-10-13 14:45:29 +02002090show pools
2091 Dump the status of internal memory pools. This is useful to track memory
2092 usage when suspecting a memory leak for example. It does exactly the same
2093 as the SIGQUIT when running in foreground except that it does not flush
2094 the pools.
2095
Willy Tarreau75c62c22018-11-22 11:02:09 +01002096show profiling
2097 Dumps the current profiling settings, one per line, as well as the command
2098 needed to change them.
2099
Willy Tarreau44aed902015-10-13 14:45:29 +02002100show servers state [<backend>]
2101 Dump the state of the servers found in the running configuration. A backend
2102 name or identifier may be provided to limit the output to this backend only.
2103
2104 The dump has the following format:
2105 - first line contains the format version (1 in this specification);
2106 - second line contains the column headers, prefixed by a sharp ('#');
2107 - third line and next ones contain data;
2108 - each line starting by a sharp ('#') is considered as a comment.
2109
Dan Lloyd8e48b872016-07-01 21:01:18 -04002110 Since multiple versions of the output may co-exist, below is the list of
Willy Tarreau44aed902015-10-13 14:45:29 +02002111 fields and their order per file format version :
2112 1:
2113 be_id: Backend unique id.
2114 be_name: Backend label.
2115 srv_id: Server unique id (in the backend).
2116 srv_name: Server label.
2117 srv_addr: Server IP address.
2118 srv_op_state: Server operational state (UP/DOWN/...).
Cyril Bonté5b2ce8a2016-11-02 00:19:58 +01002119 0 = SRV_ST_STOPPED
2120 The server is down.
2121 1 = SRV_ST_STARTING
2122 The server is warming up (up but
2123 throttled).
2124 2 = SRV_ST_RUNNING
2125 The server is fully up.
2126 3 = SRV_ST_STOPPING
2127 The server is up but soft-stopping
2128 (eg: 404).
Willy Tarreau44aed902015-10-13 14:45:29 +02002129 srv_admin_state: Server administrative state (MAINT/DRAIN/...).
Cyril Bonté5b2ce8a2016-11-02 00:19:58 +01002130 The state is actually a mask of values :
2131 0x01 = SRV_ADMF_FMAINT
2132 The server was explicitly forced into
2133 maintenance.
2134 0x02 = SRV_ADMF_IMAINT
2135 The server has inherited the maintenance
2136 status from a tracked server.
2137 0x04 = SRV_ADMF_CMAINT
2138 The server is in maintenance because of
2139 the configuration.
2140 0x08 = SRV_ADMF_FDRAIN
2141 The server was explicitly forced into
2142 drain state.
2143 0x10 = SRV_ADMF_IDRAIN
2144 The server has inherited the drain status
2145 from a tracked server.
Baptiste Assmann89aa7f32016-11-02 21:31:27 +01002146 0x20 = SRV_ADMF_RMAINT
2147 The server is in maintenance because of an
2148 IP address resolution failure.
Frédéric Lécailleb418c122017-04-26 11:24:02 +02002149 0x40 = SRV_ADMF_HMAINT
2150 The server FQDN was set from stats socket.
2151
Willy Tarreau44aed902015-10-13 14:45:29 +02002152 srv_uweight: User visible server's weight.
2153 srv_iweight: Server's initial weight.
2154 srv_time_since_last_change: Time since last operational change.
2155 srv_check_status: Last health check status.
2156 srv_check_result: Last check result (FAILED/PASSED/...).
Cyril Bonté5b2ce8a2016-11-02 00:19:58 +01002157 0 = CHK_RES_UNKNOWN
2158 Initialized to this by default.
2159 1 = CHK_RES_NEUTRAL
2160 Valid check but no status information.
2161 2 = CHK_RES_FAILED
2162 Check failed.
2163 3 = CHK_RES_PASSED
2164 Check succeeded and server is fully up
2165 again.
2166 4 = CHK_RES_CONDPASS
2167 Check reports the server doesn't want new
2168 sessions.
Willy Tarreau44aed902015-10-13 14:45:29 +02002169 srv_check_health: Checks rise / fall current counter.
2170 srv_check_state: State of the check (ENABLED/PAUSED/...).
Cyril Bonté5b2ce8a2016-11-02 00:19:58 +01002171 The state is actually a mask of values :
2172 0x01 = CHK_ST_INPROGRESS
2173 A check is currently running.
2174 0x02 = CHK_ST_CONFIGURED
2175 This check is configured and may be
2176 enabled.
2177 0x04 = CHK_ST_ENABLED
2178 This check is currently administratively
2179 enabled.
2180 0x08 = CHK_ST_PAUSED
2181 Checks are paused because of maintenance
2182 (health only).
Willy Tarreau44aed902015-10-13 14:45:29 +02002183 srv_agent_state: State of the agent check (ENABLED/PAUSED/...).
Cyril Bonté5b2ce8a2016-11-02 00:19:58 +01002184 This state uses the same mask values as
2185 "srv_check_state", adding this specific one :
2186 0x10 = CHK_ST_AGENT
2187 Check is an agent check (otherwise it's a
2188 health check).
Willy Tarreau44aed902015-10-13 14:45:29 +02002189 bk_f_forced_id: Flag to know if the backend ID is forced by
2190 configuration.
2191 srv_f_forced_id: Flag to know if the server's ID is forced by
2192 configuration.
Frédéric Lécailleb418c122017-04-26 11:24:02 +02002193 srv_fqdn: Server FQDN.
Frédéric Lécaille31694712017-08-01 08:47:19 +02002194 srv_port: Server port.
Baptiste Assmann6d0f38f2018-07-02 17:00:54 +02002195 srvrecord: DNS SRV record associated to this SRV.
Willy Tarreau44aed902015-10-13 14:45:29 +02002196
2197show sess
2198 Dump all known sessions. Avoid doing this on slow connections as this can
2199 be huge. This command is restricted and can only be issued on sockets
2200 configured for levels "operator" or "admin".
2201
2202show sess <id>
2203 Display a lot of internal information about the specified session identifier.
2204 This identifier is the first field at the beginning of the lines in the dumps
2205 of "show sess" (it corresponds to the session pointer). Those information are
2206 useless to most users but may be used by haproxy developers to troubleshoot a
2207 complex bug. The output format is intentionally not documented so that it can
2208 freely evolve depending on demands. You may find a description of all fields
2209 returned in src/dumpstats.c
2210
2211 The special id "all" dumps the states of all sessions, which must be avoided
2212 as much as possible as it is highly CPU intensive and can take a lot of time.
2213
Simon Horman05ee2132017-01-04 09:37:25 +01002214show stat [{<iid>|<proxy>} <type> <sid>] [typed|json]
2215 Dump statistics using the CSV format; using the extended typed output
2216 format described in the section above if "typed" is passed after the
2217 other arguments; or in JSON if "json" is passed after the other arguments
2218 . By passing <id>, <type> and <sid>, it is possible to dump only selected
2219 items :
Willy Tarreaua1b1ed52016-11-25 08:50:58 +01002220 - <iid> is a proxy ID, -1 to dump everything. Alternatively, a proxy name
2221 <proxy> may be specified. In this case, this proxy's ID will be used as
2222 the ID selector.
Willy Tarreau44aed902015-10-13 14:45:29 +02002223 - <type> selects the type of dumpable objects : 1 for frontends, 2 for
2224 backends, 4 for servers, -1 for everything. These values can be ORed,
2225 for example:
2226 1 + 2 = 3 -> frontend + backend.
2227 1 + 2 + 4 = 7 -> frontend + backend + server.
2228 - <sid> is a server ID, -1 to dump everything from the selected proxy.
2229
2230 Example :
2231 $ echo "show info;show stat" | socat stdio unix-connect:/tmp/sock1
2232 >>> Name: HAProxy
2233 Version: 1.4-dev2-49
2234 Release_date: 2009/09/23
2235 Nbproc: 1
2236 Process_num: 1
2237 (...)
2238
2239 # pxname,svname,qcur,qmax,scur,smax,slim,stot,bin,bout,dreq, (...)
2240 stats,FRONTEND,,,0,0,1000,0,0,0,0,0,0,,,,,OPEN,,,,,,,,,1,1,0, (...)
2241 stats,BACKEND,0,0,0,0,1000,0,0,0,0,0,,0,0,0,0,UP,0,0,0,,0,250,(...)
2242 (...)
2243 www1,BACKEND,0,0,0,0,1000,0,0,0,0,0,,0,0,0,0,UP,1,1,0,,0,250, (...)
2244
2245 $
2246
Willy Tarreau5d8b9792016-03-11 11:09:34 +01002247 In this example, two commands have been issued at once. That way it's easy to
2248 find which process the stats apply to in multi-process mode. This is not
2249 needed in the typed output format as the process number is reported on each
2250 line. Notice the empty line after the information output which marks the end
2251 of the first block. A similar empty line appears at the end of the second
2252 block (stats) so that the reader knows the output has not been truncated.
2253
2254 When "typed" is specified, the output format is more suitable to monitoring
2255 tools because it provides numeric positions and indicates the type of each
2256 output field. Each value stands on its own line with process number, element
2257 number, nature, origin and scope. This same format is available via the HTTP
2258 stats by passing ";typed" after the URI. It is very important to note that in
Dan Lloyd8e48b872016-07-01 21:01:18 -04002259 typed output format, the dump for a single object is contiguous so that there
Willy Tarreau5d8b9792016-03-11 11:09:34 +01002260 is no need for a consumer to store everything at once.
2261
2262 When using the typed output format, each line is made of 4 columns delimited
2263 by colons (':'). The first column is a dot-delimited series of 5 elements. The
2264 first element is a letter indicating the type of the object being described.
2265 At the moment the following object types are known : 'F' for a frontend, 'B'
2266 for a backend, 'L' for a listener, and 'S' for a server. The second element
2267 The second element is a positive integer representing the unique identifier of
2268 the proxy the object belongs to. It is equivalent to the "iid" column of the
2269 CSV output and matches the value in front of the optional "id" directive found
2270 in the frontend or backend section. The third element is a positive integer
2271 containing the unique object identifier inside the proxy, and corresponds to
2272 the "sid" column of the CSV output. ID 0 is reported when dumping a frontend
2273 or a backend. For a listener or a server, this corresponds to their respective
2274 ID inside the proxy. The fourth element is the numeric position of the field
2275 in the list (starting at zero). This position shall not change over time, but
2276 holes are to be expected, depending on build options or if some fields are
2277 deleted in the future. The fifth element is the field name as it appears in
2278 the CSV output. The sixth element is a positive integer and is the relative
2279 process number starting at 1.
2280
2281 The rest of the line starting after the first colon follows the "typed output
2282 format" described in the section above. In short, the second column (after the
2283 first ':') indicates the origin, nature and scope of the variable. The third
2284 column indicates the type of the field, among "s32", "s64", "u32", "u64" and
2285 "str". Then the fourth column is the value itself, which the consumer knows
2286 how to parse thanks to column 3 and how to process thanks to column 2.
2287
2288 Thus the overall line format in typed mode is :
2289
2290 <obj>.<px_id>.<id>.<fpos>.<fname>.<process_num>:<tags>:<type>:<value>
2291
2292 Here's an example of typed output format :
2293
2294 $ echo "show stat typed" | socat stdio unix-connect:/tmp/sock1
2295 F.2.0.0.pxname.1:MGP:str:private-frontend
2296 F.2.0.1.svname.1:MGP:str:FRONTEND
2297 F.2.0.8.bin.1:MGP:u64:0
2298 F.2.0.9.bout.1:MGP:u64:0
2299 F.2.0.40.hrsp_2xx.1:MGP:u64:0
2300 L.2.1.0.pxname.1:MGP:str:private-frontend
2301 L.2.1.1.svname.1:MGP:str:sock-1
2302 L.2.1.17.status.1:MGP:str:OPEN
2303 L.2.1.73.addr.1:MGP:str:0.0.0.0:8001
2304 S.3.13.60.rtime.1:MCP:u32:0
2305 S.3.13.61.ttime.1:MCP:u32:0
2306 S.3.13.62.agent_status.1:MGP:str:L4TOUT
2307 S.3.13.64.agent_duration.1:MGP:u64:2001
2308 S.3.13.65.check_desc.1:MCP:str:Layer4 timeout
2309 S.3.13.66.agent_desc.1:MCP:str:Layer4 timeout
2310 S.3.13.67.check_rise.1:MCP:u32:2
2311 S.3.13.68.check_fall.1:MCP:u32:3
2312 S.3.13.69.check_health.1:SGP:u32:0
2313 S.3.13.70.agent_rise.1:MaP:u32:1
2314 S.3.13.71.agent_fall.1:SGP:u32:1
2315 S.3.13.72.agent_health.1:SGP:u32:1
2316 S.3.13.73.addr.1:MCP:str:1.255.255.255:8888
2317 S.3.13.75.mode.1:MAP:str:http
2318 B.3.0.0.pxname.1:MGP:str:private-backend
2319 B.3.0.1.svname.1:MGP:str:BACKEND
2320 B.3.0.2.qcur.1:MGP:u32:0
2321 B.3.0.3.qmax.1:MGP:u32:0
2322 B.3.0.4.scur.1:MGP:u32:0
2323 B.3.0.5.smax.1:MGP:u32:0
2324 B.3.0.6.slim.1:MGP:u32:1000
2325 B.3.0.55.lastsess.1:MMP:s32:-1
2326 (...)
2327
Simon Horman1084a362016-11-21 17:00:24 +01002328 In the typed format, the presence of the process ID at the end of the
2329 first column makes it very easy to visually aggregate outputs from
2330 multiple processes, as show in the example below where each line appears
2331 for each process :
Willy Tarreau5d8b9792016-03-11 11:09:34 +01002332
2333 $ ( echo show stat typed | socat /var/run/haproxy.sock1 - ; \
2334 echo show stat typed | socat /var/run/haproxy.sock2 - ) | \
2335 sort -t . -k 1,1 -k 2,2n -k 3,3n -k 4,4n -k 5,5 -k 6,6n
2336 B.3.0.0.pxname.1:MGP:str:private-backend
2337 B.3.0.0.pxname.2:MGP:str:private-backend
2338 B.3.0.1.svname.1:MGP:str:BACKEND
2339 B.3.0.1.svname.2:MGP:str:BACKEND
2340 B.3.0.2.qcur.1:MGP:u32:0
2341 B.3.0.2.qcur.2:MGP:u32:0
2342 B.3.0.3.qmax.1:MGP:u32:0
2343 B.3.0.3.qmax.2:MGP:u32:0
2344 B.3.0.4.scur.1:MGP:u32:0
2345 B.3.0.4.scur.2:MGP:u32:0
2346 B.3.0.5.smax.1:MGP:u32:0
2347 B.3.0.5.smax.2:MGP:u32:0
2348 B.3.0.6.slim.1:MGP:u32:1000
2349 B.3.0.6.slim.2:MGP:u32:1000
2350 (...)
Willy Tarreau44aed902015-10-13 14:45:29 +02002351
Simon Horman05ee2132017-01-04 09:37:25 +01002352 The format of JSON output is described in a schema which may be output
Simon Horman6f6bb382017-01-04 09:37:26 +01002353 using "show schema json".
2354
2355 The JSON output contains no extra whitespace in order to reduce the
2356 volume of output. For human consumption passing the output through a
2357 pretty printer may be helpful. Example :
2358
2359 $ echo "show stat json" | socat /var/run/haproxy.sock stdio | \
2360 python -m json.tool
Simon Horman05ee2132017-01-04 09:37:25 +01002361
2362 The JSON output contains no extra whitespace in order to reduce the
2363 volume of output. For human consumption passing the output through a
2364 pretty printer may be helpful. Example :
2365
2366 $ echo "show stat json" | socat /var/run/haproxy.sock stdio | \
2367 python -m json.tool
2368
Willy Tarreau44aed902015-10-13 14:45:29 +02002369show stat resolvers [<resolvers section id>]
2370 Dump statistics for the given resolvers section, or all resolvers sections
2371 if no section is supplied.
2372
2373 For each name server, the following counters are reported:
2374 sent: number of DNS requests sent to this server
2375 valid: number of DNS valid responses received from this server
2376 update: number of DNS responses used to update the server's IP address
2377 cname: number of CNAME responses
2378 cname_error: CNAME errors encountered with this server
2379 any_err: number of empty response (IE: server does not support ANY type)
2380 nx: non existent domain response received from this server
2381 timeout: how many time this server did not answer in time
2382 refused: number of requests refused by this server
2383 other: any other DNS errors
2384 invalid: invalid DNS response (from a protocol point of view)
2385 too_big: too big response
2386 outdated: number of response arrived too late (after an other name server)
2387
2388show table
2389 Dump general information on all known stick-tables. Their name is returned
2390 (the name of the proxy which holds them), their type (currently zero, always
2391 IP), their size in maximum possible number of entries, and the number of
2392 entries currently in use.
2393
2394 Example :
2395 $ echo "show table" | socat stdio /tmp/sock1
2396 >>> # table: front_pub, type: ip, size:204800, used:171454
2397 >>> # table: back_rdp, type: ip, size:204800, used:0
2398
2399show table <name> [ data.<type> <operator> <value> ] | [ key <key> ]
2400 Dump contents of stick-table <name>. In this mode, a first line of generic
2401 information about the table is reported as with "show table", then all
2402 entries are dumped. Since this can be quite heavy, it is possible to specify
2403 a filter in order to specify what entries to display.
2404
2405 When the "data." form is used the filter applies to the stored data (see
2406 "stick-table" in section 4.2). A stored data type must be specified
2407 in <type>, and this data type must be stored in the table otherwise an
2408 error is reported. The data is compared according to <operator> with the
2409 64-bit integer <value>. Operators are the same as with the ACLs :
2410
2411 - eq : match entries whose data is equal to this value
2412 - ne : match entries whose data is not equal to this value
2413 - le : match entries whose data is less than or equal to this value
2414 - ge : match entries whose data is greater than or equal to this value
2415 - lt : match entries whose data is less than this value
2416 - gt : match entries whose data is greater than this value
2417
2418
2419 When the key form is used the entry <key> is shown. The key must be of the
2420 same type as the table, which currently is limited to IPv4, IPv6, integer,
2421 and string.
2422
2423 Example :
2424 $ echo "show table http_proxy" | socat stdio /tmp/sock1
2425 >>> # table: http_proxy, type: ip, size:204800, used:2
2426 >>> 0x80e6a4c: key=127.0.0.1 use=0 exp=3594729 gpc0=0 conn_rate(30000)=1 \
2427 bytes_out_rate(60000)=187
2428 >>> 0x80e6a80: key=127.0.0.2 use=0 exp=3594740 gpc0=1 conn_rate(30000)=10 \
2429 bytes_out_rate(60000)=191
2430
2431 $ echo "show table http_proxy data.gpc0 gt 0" | socat stdio /tmp/sock1
2432 >>> # table: http_proxy, type: ip, size:204800, used:2
2433 >>> 0x80e6a80: key=127.0.0.2 use=0 exp=3594740 gpc0=1 conn_rate(30000)=10 \
2434 bytes_out_rate(60000)=191
2435
2436 $ echo "show table http_proxy data.conn_rate gt 5" | \
2437 socat stdio /tmp/sock1
2438 >>> # table: http_proxy, type: ip, size:204800, used:2
2439 >>> 0x80e6a80: key=127.0.0.2 use=0 exp=3594740 gpc0=1 conn_rate(30000)=10 \
2440 bytes_out_rate(60000)=191
2441
2442 $ echo "show table http_proxy key 127.0.0.2" | \
2443 socat stdio /tmp/sock1
2444 >>> # table: http_proxy, type: ip, size:204800, used:2
2445 >>> 0x80e6a80: key=127.0.0.2 use=0 exp=3594740 gpc0=1 conn_rate(30000)=10 \
2446 bytes_out_rate(60000)=191
2447
2448 When the data criterion applies to a dynamic value dependent on time such as
2449 a bytes rate, the value is dynamically computed during the evaluation of the
2450 entry in order to decide whether it has to be dumped or not. This means that
2451 such a filter could match for some time then not match anymore because as
2452 time goes, the average event rate drops.
2453
2454 It is possible to use this to extract lists of IP addresses abusing the
2455 service, in order to monitor them or even blacklist them in a firewall.
2456 Example :
2457 $ echo "show table http_proxy data.gpc0 gt 0" \
2458 | socat stdio /tmp/sock1 \
2459 | fgrep 'key=' | cut -d' ' -f2 | cut -d= -f2 > abusers-ip.txt
2460 ( or | awk '/key/{ print a[split($2,a,"=")]; }' )
2461
William Lallemandbb933462016-05-31 21:09:53 +02002462show tls-keys [id|*]
2463 Dump all loaded TLS ticket keys references. The TLS ticket key reference ID
2464 and the file from which the keys have been loaded is shown. Both of those
2465 can be used to update the TLS keys using "set ssl tls-key". If an ID is
2466 specified as parameter, it will dump the tickets, using * it will dump every
2467 keys from every references.
Willy Tarreau44aed902015-10-13 14:45:29 +02002468
Simon Horman6f6bb382017-01-04 09:37:26 +01002469show schema json
2470 Dump the schema used for the output of "show info json" and "show stat json".
2471
2472 The contains no extra whitespace in order to reduce the volume of output.
2473 For human consumption passing the output through a pretty printer may be
2474 helpful. Example :
2475
2476 $ echo "show schema json" | socat /var/run/haproxy.sock stdio | \
2477 python -m json.tool
2478
2479 The schema follows "JSON Schema" (json-schema.org) and accordingly
2480 verifiers may be used to verify the output of "show info json" and "show
2481 stat json" against the schema.
2482
2483
Willy Tarreau44aed902015-10-13 14:45:29 +02002484shutdown frontend <frontend>
2485 Completely delete the specified frontend. All the ports it was bound to will
2486 be released. It will not be possible to enable the frontend anymore after
2487 this operation. This is intended to be used in environments where stopping a
2488 proxy is not even imaginable but a misconfigured proxy must be fixed. That
2489 way it's possible to release the port and bind it into another process to
2490 restore operations. The frontend will not appear at all on the stats page
2491 once it is terminated.
2492
2493 The frontend may be specified either by its name or by its numeric ID,
2494 prefixed with a sharp ('#').
2495
2496 This command is restricted and can only be issued on sockets configured for
2497 level "admin".
2498
2499shutdown session <id>
2500 Immediately terminate the session matching the specified session identifier.
2501 This identifier is the first field at the beginning of the lines in the dumps
2502 of "show sess" (it corresponds to the session pointer). This can be used to
2503 terminate a long-running session without waiting for a timeout or when an
2504 endless transfer is ongoing. Such terminated sessions are reported with a 'K'
2505 flag in the logs.
2506
2507shutdown sessions server <backend>/<server>
2508 Immediately terminate all the sessions attached to the specified server. This
2509 can be used to terminate long-running sessions after a server is put into
2510 maintenance mode, for instance. Such terminated sessions are reported with a
2511 'K' flag in the logs.
2512
Willy Tarreau2212e6a2015-10-13 14:40:55 +02002513
William Lallemand142db372018-12-11 18:56:45 +010025149.4. Master CLI
2515---------------
2516
2517The master CLI is a socket bound to the master process in master-worker mode.
2518This CLI gives access to the unix socket commands in every running or leaving
2519processes and allows a basic supervision of those processes.
2520
2521The master CLI is configurable only from the haproxy program arguments with
2522the -S option. This option also takes bind options separated by commas.
2523
2524Example:
2525
2526 # haproxy -W -S 127.0.0.1:1234 -f test1.cfg
2527 # haproxy -Ws -S /tmp/master-socket,uid,1000,gid,1000,mode,600 -f test1.cfg
William Lallemandb7ea1412018-12-13 09:05:47 +01002528 # haproxy -W -S /tmp/master-socket,level,user -f test1.cfg
William Lallemand142db372018-12-11 18:56:45 +01002529
2530The master CLI introduces a new 'show proc' command to surpervise the
2531processes:
2532
2533Example:
2534
2535 $ echo 'show proc' | socat /var/run/haproxy-master.sock -
2536 #<PID> <type> <relative PID> <reloads> <uptime>
2537 1162 master 0 5 0d 00h02m07s
2538 # workers
2539 1271 worker 1 0 0d 00h00m00s
2540 1272 worker 2 0 0d 00h00m00s
2541 # old workers
William Lallemand256bf0d2018-12-12 13:45:57 +01002542 1233 worker [was: 1] 3 0d 00h00m43s
William Lallemand142db372018-12-11 18:56:45 +01002543
2544
2545In this example, the master has been reloaded 5 times but one of the old
2546worker is still running and survived 3 reloads. You could access the CLI of
2547this worker to understand what's going on.
2548
Willy Tarreau52880f92018-12-15 13:30:03 +01002549When the prompt is enabled (via the "prompt" command), the context the CLI is
2550working on is displayed in the prompt. The master is identified by the "master"
2551string, and other processes are identified with their PID. In case the last
2552reload failed, the master prompt will be changed to "master[ReloadFailed]>" so
2553that it becomes visible that the process is still running on the previous
2554configuration and that the new configuration is not operational.
2555
William Lallemand142db372018-12-11 18:56:45 +01002556The master CLI uses a special prefix notation to access the multiple
2557processes. This notation is easily identifiable as it begins by a @.
2558
2559A @ prefix can be followed by a relative process number or by an exclamation
2560point and a PID. (e.g. @1 or @!1271). A @ alone could be use to specify the
2561master. Leaving processes are only accessible with the PID as relative process
2562number are only usable with the current processes.
2563
2564Examples:
2565
2566 $ socat /var/run/haproxy-master.sock readline
2567 prompt
2568 master> @1 show info; @2 show info
2569 [...]
2570 Process_num: 1
2571 Pid: 1271
2572 [...]
2573 Process_num: 2
2574 Pid: 1272
2575 [...]
2576 master>
2577
2578 $ echo '@!1271 show info; @!1272 show info' | socat /var/run/haproxy-master.sock -
2579 [...]
2580
2581A prefix could be use as a command, which will send every next commands to
2582the specified process.
2583
2584Examples:
2585
2586 $ socat /var/run/haproxy-master.sock readline
2587 prompt
2588 master> @1
2589 1271> show info
2590 [...]
2591 1271> show stat
2592 [...]
2593 1271> @
2594 master>
2595
2596 $ echo '@1; show info; show stat; @2; show info; show stat' | socat /var/run/haproxy-master.sock -
2597 [...]
2598
William Lallemanda57b7e32018-12-14 21:11:31 +01002599You can also reload the HAProxy master process with the "reload" command which
2600does the same as a `kill -USR2` on the master process, provided that the user
2601has at least "operator" or "admin" privileges.
2602
2603Example:
2604
2605 $ echo "reload" | socat /var/run/haproxy-master.sock
2606
2607Note that a reload will close the connection to the master CLI.
2608
William Lallemand142db372018-12-11 18:56:45 +01002609
Willy Tarreau2212e6a2015-10-13 14:40:55 +0200261010. Tricks for easier configuration management
2611----------------------------------------------
2612
2613It is very common that two HAProxy nodes constituting a cluster share exactly
2614the same configuration modulo a few addresses. Instead of having to maintain a
2615duplicate configuration for each node, which will inevitably diverge, it is
2616possible to include environment variables in the configuration. Thus multiple
2617configuration may share the exact same file with only a few different system
2618wide environment variables. This started in version 1.5 where only addresses
2619were allowed to include environment variables, and 1.6 goes further by
2620supporting environment variables everywhere. The syntax is the same as in the
2621UNIX shell, a variable starts with a dollar sign ('$'), followed by an opening
2622curly brace ('{'), then the variable name followed by the closing brace ('}').
2623Except for addresses, environment variables are only interpreted in arguments
2624surrounded with double quotes (this was necessary not to break existing setups
2625using regular expressions involving the dollar symbol).
2626
2627Environment variables also make it convenient to write configurations which are
2628expected to work on various sites where only the address changes. It can also
2629permit to remove passwords from some configs. Example below where the the file
2630"site1.env" file is sourced by the init script upon startup :
2631
2632 $ cat site1.env
2633 LISTEN=192.168.1.1
2634 CACHE_PFX=192.168.11
2635 SERVER_PFX=192.168.22
2636 LOGGER=192.168.33.1
2637 STATSLP=admin:pa$$w0rd
2638 ABUSERS=/etc/haproxy/abuse.lst
2639 TIMEOUT=10s
2640
2641 $ cat haproxy.cfg
2642 global
2643 log "${LOGGER}:514" local0
2644
2645 defaults
2646 mode http
2647 timeout client "${TIMEOUT}"
2648 timeout server "${TIMEOUT}"
2649 timeout connect 5s
2650
2651 frontend public
2652 bind "${LISTEN}:80"
2653 http-request reject if { src -f "${ABUSERS}" }
2654 stats uri /stats
2655 stats auth "${STATSLP}"
2656 use_backend cache if { path_end .jpg .css .ico }
2657 default_backend server
2658
2659 backend cache
2660 server cache1 "${CACHE_PFX}.1:18080" check
2661 server cache2 "${CACHE_PFX}.2:18080" check
2662
2663 backend server
2664 server cache1 "${SERVER_PFX}.1:8080" check
2665 server cache2 "${SERVER_PFX}.2:8080" check
2666
2667
266811. Well-known traps to avoid
2669-----------------------------
2670
2671Once in a while, someone reports that after a system reboot, the haproxy
2672service wasn't started, and that once they start it by hand it works. Most
2673often, these people are running a clustered IP address mechanism such as
2674keepalived, to assign the service IP address to the master node only, and while
2675it used to work when they used to bind haproxy to address 0.0.0.0, it stopped
2676working after they bound it to the virtual IP address. What happens here is
2677that when the service starts, the virtual IP address is not yet owned by the
2678local node, so when HAProxy wants to bind to it, the system rejects this
2679because it is not a local IP address. The fix doesn't consist in delaying the
2680haproxy service startup (since it wouldn't stand a restart), but instead to
2681properly configure the system to allow binding to non-local addresses. This is
2682easily done on Linux by setting the net.ipv4.ip_nonlocal_bind sysctl to 1. This
2683is also needed in order to transparently intercept the IP traffic that passes
2684through HAProxy for a specific target address.
2685
2686Multi-process configurations involving source port ranges may apparently seem
2687to work but they will cause some random failures under high loads because more
2688than one process may try to use the same source port to connect to the same
2689server, which is not possible. The system will report an error and a retry will
2690happen, picking another port. A high value in the "retries" parameter may hide
2691the effect to a certain extent but this also comes with increased CPU usage and
2692processing time. Logs will also report a certain number of retries. For this
2693reason, port ranges should be avoided in multi-process configurations.
2694
Dan Lloyd8e48b872016-07-01 21:01:18 -04002695Since HAProxy uses SO_REUSEPORT and supports having multiple independent
Willy Tarreau2212e6a2015-10-13 14:40:55 +02002696processes bound to the same IP:port, during troubleshooting it can happen that
2697an old process was not stopped before a new one was started. This provides
2698absurd test results which tend to indicate that any change to the configuration
2699is ignored. The reason is that in fact even the new process is restarted with a
2700new configuration, the old one also gets some incoming connections and
2701processes them, returning unexpected results. When in doubt, just stop the new
2702process and try again. If it still works, it very likely means that an old
2703process remains alive and has to be stopped. Linux's "netstat -lntp" is of good
2704help here.
2705
2706When adding entries to an ACL from the command line (eg: when blacklisting a
2707source address), it is important to keep in mind that these entries are not
2708synchronized to the file and that if someone reloads the configuration, these
2709updates will be lost. While this is often the desired effect (for blacklisting)
2710it may not necessarily match expectations when the change was made as a fix for
2711a problem. See the "add acl" action of the CLI interface.
2712
2713
271412. Debugging and performance issues
2715------------------------------------
2716
2717When HAProxy is started with the "-d" option, it will stay in the foreground
2718and will print one line per event, such as an incoming connection, the end of a
2719connection, and for each request or response header line seen. This debug
2720output is emitted before the contents are processed, so they don't consider the
2721local modifications. The main use is to show the request and response without
2722having to run a network sniffer. The output is less readable when multiple
2723connections are handled in parallel, though the "debug2ansi" and "debug2html"
2724scripts found in the examples/ directory definitely help here by coloring the
2725output.
2726
2727If a request or response is rejected because HAProxy finds it is malformed, the
2728best thing to do is to connect to the CLI and issue "show errors", which will
2729report the last captured faulty request and response for each frontend and
2730backend, with all the necessary information to indicate precisely the first
2731character of the input stream that was rejected. This is sometimes needed to
2732prove to customers or to developers that a bug is present in their code. In
2733this case it is often possible to relax the checks (but still keep the
2734captures) using "option accept-invalid-http-request" or its equivalent for
2735responses coming from the server "option accept-invalid-http-response". Please
2736see the configuration manual for more details.
2737
2738Example :
2739
2740 > show errors
2741 Total events captured on [13/Oct/2015:13:43:47.169] : 1
2742
2743 [13/Oct/2015:13:43:40.918] frontend HAProxyLocalStats (#2): invalid request
2744 backend <NONE> (#-1), server <NONE> (#-1), event #0
2745 src 127.0.0.1:51981, session #0, session flags 0x00000080
2746 HTTP msg state 26, msg flags 0x00000000, tx flags 0x00000000
2747 HTTP chunk len 0 bytes, HTTP body len 0 bytes
2748 buffer flags 0x00808002, out 0 bytes, total 31 bytes
2749 pending 31 bytes, wrapping at 8040, error at position 13:
2750
2751 00000 GET /invalid request HTTP/1.1\r\n
2752
2753
2754The output of "show info" on the CLI provides a number of useful information
2755regarding the maximum connection rate ever reached, maximum SSL key rate ever
2756reached, and in general all information which can help to explain temporary
2757issues regarding CPU or memory usage. Example :
2758
2759 > show info
2760 Name: HAProxy
2761 Version: 1.6-dev7-e32d18-17
2762 Release_date: 2015/10/12
2763 Nbproc: 1
2764 Process_num: 1
2765 Pid: 7949
2766 Uptime: 0d 0h02m39s
2767 Uptime_sec: 159
2768 Memmax_MB: 0
2769 Ulimit-n: 120032
2770 Maxsock: 120032
2771 Maxconn: 60000
2772 Hard_maxconn: 60000
2773 CurrConns: 0
2774 CumConns: 3
2775 CumReq: 3
2776 MaxSslConns: 0
2777 CurrSslConns: 0
2778 CumSslConns: 0
2779 Maxpipes: 0
2780 PipesUsed: 0
2781 PipesFree: 0
2782 ConnRate: 0
2783 ConnRateLimit: 0
2784 MaxConnRate: 1
2785 SessRate: 0
2786 SessRateLimit: 0
2787 MaxSessRate: 1
2788 SslRate: 0
2789 SslRateLimit: 0
2790 MaxSslRate: 0
2791 SslFrontendKeyRate: 0
2792 SslFrontendMaxKeyRate: 0
2793 SslFrontendSessionReuse_pct: 0
2794 SslBackendKeyRate: 0
2795 SslBackendMaxKeyRate: 0
2796 SslCacheLookups: 0
2797 SslCacheMisses: 0
2798 CompressBpsIn: 0
2799 CompressBpsOut: 0
2800 CompressBpsRateLim: 0
2801 ZlibMemUsage: 0
2802 MaxZlibMemUsage: 0
2803 Tasks: 5
2804 Run_queue: 1
2805 Idle_pct: 100
2806 node: wtap
2807 description:
2808
2809When an issue seems to randomly appear on a new version of HAProxy (eg: every
2810second request is aborted, occasional crash, etc), it is worth trying to enable
Dan Lloyd8e48b872016-07-01 21:01:18 -04002811memory poisoning so that each call to malloc() is immediately followed by the
Willy Tarreau2212e6a2015-10-13 14:40:55 +02002812filling of the memory area with a configurable byte. By default this byte is
28130x50 (ASCII for 'P'), but any other byte can be used, including zero (which
2814will have the same effect as a calloc() and which may make issues disappear).
Dan Lloyd8e48b872016-07-01 21:01:18 -04002815Memory poisoning is enabled on the command line using the "-dM" option. It
Willy Tarreau2212e6a2015-10-13 14:40:55 +02002816slightly hurts performance and is not recommended for use in production. If
Dan Lloyd8e48b872016-07-01 21:01:18 -04002817an issue happens all the time with it or never happens when poisoning uses
Willy Tarreau2212e6a2015-10-13 14:40:55 +02002818byte zero, it clearly means you've found a bug and you definitely need to
2819report it. Otherwise if there's no clear change, the problem it is not related.
2820
2821When debugging some latency issues, it is important to use both strace and
2822tcpdump on the local machine, and another tcpdump on the remote system. The
2823reason for this is that there are delays everywhere in the processing chain and
2824it is important to know which one is causing latency to know where to act. In
2825practice, the local tcpdump will indicate when the input data come in. Strace
2826will indicate when haproxy receives these data (using recv/recvfrom). Warning,
2827openssl uses read()/write() syscalls instead of recv()/send(). Strace will also
2828show when haproxy sends the data, and tcpdump will show when the system sends
2829these data to the interface. Then the external tcpdump will show when the data
2830sent are really received (since the local one only shows when the packets are
2831queued). The benefit of sniffing on the local system is that strace and tcpdump
2832will use the same reference clock. Strace should be used with "-tts200" to get
2833complete timestamps and report large enough chunks of data to read them.
2834Tcpdump should be used with "-nvvttSs0" to report full packets, real sequence
2835numbers and complete timestamps.
2836
2837In practice, received data are almost always immediately received by haproxy
2838(unless the machine has a saturated CPU or these data are invalid and not
2839delivered). If these data are received but not sent, it generally is because
2840the output buffer is saturated (ie: recipient doesn't consume the data fast
2841enough). This can be confirmed by seeing that the polling doesn't notify of
2842the ability to write on the output file descriptor for some time (it's often
2843easier to spot in the strace output when the data finally leave and then roll
2844back to see when the write event was notified). It generally matches an ACK
2845received from the recipient, and detected by tcpdump. Once the data are sent,
2846they may spend some time in the system doing nothing. Here again, the TCP
2847congestion window may be limited and not allow these data to leave, waiting for
2848an ACK to open the window. If the traffic is idle and the data take 40 ms or
2849200 ms to leave, it's a different issue (which is not an issue), it's the fact
2850that the Nagle algorithm prevents empty packets from leaving immediately, in
2851hope that they will be merged with subsequent data. HAProxy automatically
2852disables Nagle in pure TCP mode and in tunnels. However it definitely remains
2853enabled when forwarding an HTTP body (and this contributes to the performance
2854improvement there by reducing the number of packets). Some HTTP non-compliant
2855applications may be sensitive to the latency when delivering incomplete HTTP
2856response messages. In this case you will have to enable "option http-no-delay"
2857to disable Nagle in order to work around their design, keeping in mind that any
2858other proxy in the chain may similarly be impacted. If tcpdump reports that data
2859leave immediately but the other end doesn't see them quickly, it can mean there
Dan Lloyd8e48b872016-07-01 21:01:18 -04002860is a congested WAN link, a congested LAN with flow control enabled and
Willy Tarreau2212e6a2015-10-13 14:40:55 +02002861preventing the data from leaving, or more commonly that HAProxy is in fact
2862running in a virtual machine and that for whatever reason the hypervisor has
2863decided that the data didn't need to be sent immediately. In virtualized
2864environments, latency issues are almost always caused by the virtualization
2865layer, so in order to save time, it's worth first comparing tcpdump in the VM
2866and on the external components. Any difference has to be credited to the
2867hypervisor and its accompanying drivers.
2868
2869When some TCP SACK segments are seen in tcpdump traces (using -vv), it always
2870means that the side sending them has got the proof of a lost packet. While not
2871seeing them doesn't mean there are no losses, seeing them definitely means the
2872network is lossy. Losses are normal on a network, but at a rate where SACKs are
2873not noticeable at the naked eye. If they appear a lot in the traces, it is
2874worth investigating exactly what happens and where the packets are lost. HTTP
2875doesn't cope well with TCP losses, which introduce huge latencies.
2876
2877The "netstat -i" command will report statistics per interface. An interface
2878where the Rx-Ovr counter grows indicates that the system doesn't have enough
2879resources to receive all incoming packets and that they're lost before being
2880processed by the network driver. Rx-Drp indicates that some received packets
2881were lost in the network stack because the application doesn't process them
2882fast enough. This can happen during some attacks as well. Tx-Drp means that
2883the output queues were full and packets had to be dropped. When using TCP it
Dan Lloyd8e48b872016-07-01 21:01:18 -04002884should be very rare, but will possibly indicate a saturated outgoing link.
Willy Tarreau2212e6a2015-10-13 14:40:55 +02002885
2886
288713. Security considerations
2888---------------------------
2889
2890HAProxy is designed to run with very limited privileges. The standard way to
2891use it is to isolate it into a chroot jail and to drop its privileges to a
2892non-root user without any permissions inside this jail so that if any future
2893vulnerability were to be discovered, its compromise would not affect the rest
2894of the system.
2895
Dan Lloyd8e48b872016-07-01 21:01:18 -04002896In order to perform a chroot, it first needs to be started as a root user. It is
Willy Tarreau2212e6a2015-10-13 14:40:55 +02002897pointless to build hand-made chroots to start the process there, these ones are
2898painful to build, are never properly maintained and always contain way more
2899bugs than the main file-system. And in case of compromise, the intruder can use
2900the purposely built file-system. Unfortunately many administrators confuse
2901"start as root" and "run as root", resulting in the uid change to be done prior
2902to starting haproxy, and reducing the effective security restrictions.
2903
2904HAProxy will need to be started as root in order to :
2905 - adjust the file descriptor limits
2906 - bind to privileged port numbers
2907 - bind to a specific network interface
2908 - transparently listen to a foreign address
2909 - isolate itself inside the chroot jail
2910 - drop to another non-privileged UID
2911
2912HAProxy may require to be run as root in order to :
2913 - bind to an interface for outgoing connections
2914 - bind to privileged source ports for outgoing connections
Dan Lloyd8e48b872016-07-01 21:01:18 -04002915 - transparently bind to a foreign address for outgoing connections
Willy Tarreau2212e6a2015-10-13 14:40:55 +02002916
2917Most users will never need the "run as root" case. But the "start as root"
2918covers most usages.
2919
2920A safe configuration will have :
2921
2922 - a chroot statement pointing to an empty location without any access
2923 permissions. This can be prepared this way on the UNIX command line :
2924
2925 # mkdir /var/empty && chmod 0 /var/empty || echo "Failed"
2926
2927 and referenced like this in the HAProxy configuration's global section :
2928
2929 chroot /var/empty
2930
2931 - both a uid/user and gid/group statements in the global section :
2932
2933 user haproxy
2934 group haproxy
2935
2936 - a stats socket whose mode, uid and gid are set to match the user and/or
2937 group allowed to access the CLI so that nobody may access it :
2938
2939 stats socket /var/run/haproxy.stat uid hatop gid hatop mode 600
2940